By OpenAISourceGitHub

Manuscripts from openai/math

Fork
Clone
git clone https://openversions.com/git/index/openai-math.git
Cite
Version link
https://openversions.com/index/openai-math/commit/dfd8c46870b48511e697cbb43a3c8c91d375fa2f
Version identifier (commit hash)
dfd8c46870b48511e697cbb43a3c8c91d375fa2f

Fork to create your own copy with credit to the source. Clone to get a Git copy on your computer.

All papers

Claims

Findings

Nonuniqueness for the periodic hard-sphere Boltzmann equation

There is a nonnegative density F0F_0 on T3×R3\mathbb{T}^3 \times \mathbb{R}^3, with bounded velocity support and ∫T3×R3F0(1+∣v∣2+∣log⁡F0∣) dx dv<∞\int_{\mathbb{T}^3 \times \mathbb{R}^3} F_0(1 + |v|^2 + |\log F_0|)\,\mathrm{d}x\,\mathrm{d}v < \infty, for which the hard-sphere Boltzmann equation has two distinct global admissible solutions, in the sense of Definition 2.1, with initial datum F0F_0. On some common initial interval both solutions are strongly continuous in L1L^1, and their collision gains and losses are integrable in space-time-velocity.

For the integrable velocity weight H(v)=e−∣v∣2⟨v⟩−m(1+∣vˉ∣−q)H(v) = e^{-|v|^2}\langle v \rangle^{-m}(1 + |\bar{v}|^{-q}) with 1<q<21 < q < 2 and m=100m = 100, the normalized Carleman gain kernel satisfies P(c,H)(v)≤Cq,m,LH(v)P(c, H)(v) \le C_{q,m,L} H(v) almost everywhere uniformly for all c∈SLc \in \mathcal{S}_L (with uniform tightness, axial non-atomicity, and transverse-width estimates), and the quadratic hot-hot gain satisfies P(H,H)≤Cq,mHP(H, H) \le C_{q,m} H almost everywhere.

The cold-jet datum F0F_0 in (4.4) has bounded velocity support and finite mass, energy, and absolute entropy, and for every finite approximation on 0≤t≤T‾0 \le t \le \overline{T}, each cold color is bounded by its free transport plus a small Maxwellian remainder, with escaped tails bounded by CAtAe−4∣v∣2C_A t^A e^{-4|v|^2} and the mixed-cold source bounded by CAtA−1H(v)C_A t^{A-1} H(v) uniformly in NN and bb.

There exists a fixed T0>0T_0 > 0 and L0<∞L_0 < \infty such that with zero hot seed (λ=0\lambda = 0) the finite approximations exist on [0,T0][0, T_0] and satisfy 0≤h≤t2H0 \le h \le t^2 H, while for every 0<T≤T00 < T \le T_0 and every NN, some λ∈[0,L0]\lambda \in [0, L_0] gives a cap-shot approximation on [0,T][0, T] satisfying 0≤h(t,x,v)≤(t+sN)H(v)0 \le h(t,x,v) \le (t+s_N)H(v) with max⁡0≤t≤T∥h(t)∥H/(t+sN)=1\max_{0 \le t \le T} \|h(t)\|_H/(t+s_N) = 1.

For any fixed T∈(0,T0]T \in (0, T_0], both the zero-seed approximations and any sequence of cap shots on [0,T][0, T] have subsequences converging to admissible solutions F∈C([0,T];L1(Ω))F \in C([0, T]; L^1(\Omega)) with Q±(F,F)∈L1([0,T]×Ω)Q^\pm(F, F) \in L^1([0, T] \times \Omega) and the common initial datum F0F_0, and each such limit has an admissible global continuation in the sense of Definition 2.1.

Under the shrinking-horizon dilation (7.1) at a cap time t∗,Nt_{*,N} with rN=t∗,N+sNr_N = t_{*,N} + s_N, any sequence of cap shots has a subsequence for which hrN⇀gh_{r_N} \rightharpoonup g in local spacetime L1L^1 on (0,U)×Rx3×Rv3(0, U) \times \mathbb{R}^3_x \times \mathbb{R}^3_v, where 0≤g(t,x,v)≤(t+l)H(v)0 \le g(t,x,v) \le (t+l)H(v) and gg solves the linear Boltzmann equation Dtg=2P(B,g)−gνBD_t g = 2P(B, g) - g\nu_B in the limiting two-sign background measure BB.

For any sequence of cap shots with Tn→0T_n \to 0, Nn→∞N_n \to \infty, and sNn/Tn→0s_{N_n}/T_n \to 0, any dilated limit gg supplied by Proposition 7.1 is nonzero on (0,U)×Rx3×Rv3(0, U) \times \mathbb{R}^3_x \times \mathbb{R}^3_v.

There are fixed positive choices of η,α,γ,K\eta, \alpha, \gamma, K (chosen before s0,d0,T0s_0, d_0, T_0) for which every dilated limit gg in Propositions 7.1 and 8.1 has a finite dilated horizon U<∞U < \infty.

As T↓0T \downarrow 0, the dilated dormant limit T−1Fd(Tt,Tx,v)→0T^{-1} F^{\mathrm{d}}(Tt, Tx, v) \to 0 in distributions on compact subsets of 0<t<10 < t < 1, x∈R3x \in \mathbb{R}^3, v∈R3v \in \mathbb{R}^3, vˉ≠0\bar{v} \neq 0.

Methods

Colored decomposition of the hard-sphere Boltzmann equation into cold jet components cic_i (retaining color only in same-color collisions) and a hot component hh (collecting all cross-color and hot collisions).

Source excerpts

“The cold components are ci, their sum is c, and h denotes the hot component. Particles retain a cold label only in collisions within that label. All other collisions create hot particles.” · Show in paper

Exact Carleman hyperplane representation of the symmetric bilinear gain P(c,h)P(c, h) and normalized kernel measures Kc\mathcal{K}_c with integrable velocity weight H(v)=e−∣v∣2⟨v⟩−m(1+∣vˉ∣−q)H(v) = e^{-|v|^2}\langle v \rangle^{-m}(1 + |\bar{v}|^{-q}).

Source excerpts

“For the normalization in (2.1), the symmetric bilinear gain is P(c, h)(v) = 2 R R3 c(p) dp R R3 δ((v − p) · (ξ − v))h(ξ) dξ.” · Show in paper

“For such a sheet and v̄ ̸= 0, define its normalized kernel by Kc(v; dp dξ) = 2 c( dp) δ((v − p) · (ξ − v)) H(ξ)/H(v) dξ.” · Show in paper

Transported Maxwellian barrier mi(t,x,v)m_i(t, x, v) and nonlinear comparison along characteristics for uniform cold-jet control.

Source excerpts

“Choose csp = 1/8 and a fixed sufficiently large Cm, depending only on J, and define, for t > 0, mi(t, x, v) = Cm K/sj d−3sj et exp[ − (2 − t)|wi|2 / d2sj − csp(D2y,i + D2z,i) / (d2sj t) ].” · Show in paper

Continuity/open-closed parameter selection of a vanishing hot seed λsNχ(x/sN,v)\lambda s_N \chi(x/s_N, v) achieving a weighted cap max⁡0≤t≤T∥h(t)∥H/(t+sN)=1\max_{0 \le t \le T} \|h(t)\|_H/(t+s_N) = 1.

Source excerpts

“For fixed N, bN, T, let O be the set of seed parameters in [0, L0] whose solution exists up to T and obeys a strict cap there. Continuous dependence and continuation make O relatively open.” · Show in paper

Uniformly integrable velocity averaging (Lemma 6.2) and weak L1L^1 compactness to pass to fixed-horizon admissible limits and concatenate with DiPerna-Lions global continuation.

Source excerpts

“Suppose that fn and gn are locally uniformly integrable in L1 on an open time-space-velocity region, and (∂t + v · ∇x)fn = gn there in distributions. Then for every smooth compactly supported velocity test ψ, the averages R ψ(v)fn(t, x, v) dv are relatively compact in local L1t,x on regions where the hypotheses hold for velocities in a neighborhood of supp ψ.” · Show in paper

Shrinking-horizon space-time dilation about a cap time t∗,Nt_{*,N} and weak-* stripe averaging to a linear Boltzmann equation in a two-sign measure background BB.

Source excerpts

“rN = t∗,N + sN, hrN(t, x, v) = r−1N hN(rN t, rN x, v), CrN(t, x, v) = rN cN(rN t, rN x, v)” · Show in paper

Iterated-gain regularization: two-gain axial mismatch/non-atomicity analysis and five-gain nonaxial rank-seven submersion with redundant plane parameters and uniform C1C^1 pullback.

Source excerpts

“The map Φ : (t1, . . . , t5, L1, . . . , L5) 7−→ (t5, x5, v5) is real analytic on the connected open set (0, ∞)5 × R15, and has rank seven almost everywhere.” · Show in paper

Jump-process Feynman-Kac (many-to-one) representation of branching collision histories, three-collision Doeblin minorization, and geometric box-transport growth to bound the dilated horizon U<∞U < \infty.

Source excerpts

“The reference history sum restricted to velocities in E is represented by a Markov flight with velocity jump rate λ(v) = 2p(v)ν0(v), jump law J(v, dv′)1E(v′)/p(v), and multiplicative weight exp(R A0 W(Vs) ds), W(v) = (2p(v) − 1)ν0(v).” · Show in paper