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

Thermal FK–Ising interfaces and massive SLE

For every bounded simply connected marked domain (Ω,a,b)(\Omega, a, b) and real mass m≠0m \neq 0, there is a curve law μΩ,a,b,m\mu_{\Omega,a,b,m} such that for every sequence of finite simply connected rhombic domains Ωδ\Omega^\delta on uniformly angle-bounded isoradial lattices converging in the marked Carath\'eodory sense to (Ω,a,b)(\Omega, a, b) (even with diverging diameters) and any nome sequence rδ∈(0,1)r_\delta \in (0,1) with rδ/δ→∣m∣/2r_\delta/\delta \to |m|/2, the normalized interfaces χδ∘γδ\chi_\delta \circ \gamma_\delta converge in law along the full sequence in the uniform curve metric dcurvd_{\mathrm{curv}} to χ∗μΩ,a,b,m\chi_* \mu_{\Omega,a,b,m}, independent of lattices, admissible domain approximations, and nome sequence; for m>0m > 0 it is the unique law in Definition 2.3, and for m<0m < 0 it is the reversal of the positive-mass law in (Ω,b,a)(\Omega, b, a) with mass ∣m∣|m| and primal/dual boundary labels interchanged.

The scalar boundary value problem Δh=−4m∣∇h∣\Delta h = -4m|\nabla h| in OO with h=1h=1 on the negative real bank and h=0h=0 on the positive real bank has a unique bounded classical solution, which satisfies 0≤h≤10 \leq h \leq 1 and depends continuously (along with its first derivatives) on the marked domain in the Carath\'eodory topology on interior compact sets when the domains lie in a fixed bounded physical region.

Under a common physical diameter bound, the laws of the interfaces transferred by their normalized maps to D‾\overline{\mathbb{D}} are tight in the oriented curve metric, every subsequential limit is a chordal Loewner trace with a continuous driving function and strict capacity growth reaching the target only at the end, and along a further coupling the curves and their half-plane drivers converge jointly.

Every subsequential curve law of Proposition 3.7 has a joint realization with a complex field Mt(w)M_t(w) on OtO_t such that Mt(w)2=Dht(w)M_t(w)^2 = D h_t(w) and ∂ˉMt+imMt‾=0\bar{\partial} M_t + i m \overline{M_t} = 0, which is locally continuous in space and time before swallowing, spatially smooth, and a continuous local martingale at each fixed point before its swallowing time in the joint natural filtration.

Under a common physical diameter bound, for any subsequential interface law of Proposition 3.7, the moving-mark energy Jt=∫Hμt(v)∣v∣2dA(v)J_t = \int_{\mathbb{H}} \frac{\mu_t(v)}{|v|^2} dA(v) satisfies E∫0∞Jt2dt≤Km2area(Ω)<∞\mathbb{E} \int_0^\infty J_t^2 dt \leq K m^2 \mathrm{area}(\Omega) < \infty, and JtJ_t is finite for almost every capacity time, almost surely.

Under a common physical diameter bound, the subsequential driving process WtW_t is a continuous semimartingale in the joint filtration satisfying dWt=16/3dBt+2π3CtdtdW_t = \sqrt{16/3} dB_t + \frac{2\pi}{3} C_t dt with ∫0TCt2dt<∞\int_0^T C_t^2 dt < \infty almost surely for every T<∞T < \infty, where BB is a Brownian motion and Ct=4π∫HP(v)μt(v)∣∇(ht∘ψt)(v)∣dA(v)C_t = \frac{4}{\pi} \int_{\mathbb{H}} P(v) \mu_t(v) |\nabla(h_t \circ \psi_t)(v)| dA(v).

Definition 2.3 determines exactly one law, and every admissible approximation whose physical domains have a common diameter bound converges to this law in the disc curve topology.

From every admissible approximation sequence (without a common diameter bound) one can extract a further subsequence and perform finite edge revelations with success probability tending to one, such that on success the unqueried connecting domain is a Dobrushin polygon on the original mesh with diameter at most 2S+202S+20 and the ordinary Dobrushin FK law, every deterministic sequence of successful outputs converges in the marked Carath\'eodory sense to (Ω,a,b)(\Omega, a, b), and normalized coordinates on the retained domain are uniformly o(1)o(1) from the original ones.

For m>0m > 0, every approximation sequence in Theorem 2.1 (without a diameter assumption) converges to the law of Definition 2.3 in the normalized-disc curve topology.

Methods

Bounded-drift Girsanov diffusion and optional stopping to prove uniqueness of bounded classical solutions to the massive scalar boundary value problem Δh=−4m∣∇h∣\Delta h = -4m|\nabla h| in simply connected domains.

Source excerpts

“Extend \beta boundedly and measurably outside O and use the diffusion dX = \sqrt{2} dB + \beta(X) dt, killed on exit from O. Bounded-drift Girsanov constructs this diffusion and makes its law equivalent to Brownian law on every finite time interval.” · Show in paper

Simultaneous union-of-regions arm comparison via boundary-sector copy identification and random-cluster monotonicity to bound unforced crossings without paying a multiplicity factor for the number of components.

Source excerpts

“Work in the graph of color c, keeping boundary-sector copies until the comparison is made. ... Identify copies that project to the same ambient lattice vertex. For q = 2 this is an increase in boundary identification, hence it increases open-edge connectivity. ... The construction couples the entire union at once and therefore pays no multiplicity factor.” · Show in paper

Predictable decision-index stopping to lift convergence of fermion squares to signed local martingales without assuming a continuous choice of sign.

Source excerpts

“Stop immediately before the first decision at which either hypothetical next square changes by more than \varepsilon, or either hypothetical next domain loses the smaller protection buffer, while the prevalue remains in the larger one. This is a predictable decision index: no future continuation is used to test the two hypothetical domains.” · Show in paper

Point-adapted Loewner clock with small-power angular contraction and dyadic shell summation over physical area to prove moving-mark finite energy E∫0∞Jt2dt<∞\mathbb{E} \int_0^\infty J_t^2 dt < \infty without assuming the driver is a semimartingale.

Source excerpts

“We prove the proposition by following one physical point in a clock adapted to its distance from the driver. Keeping a small positive power of its angular coordinate is necessary when summing over physical area.” · Show in paper

Height-cutoff stochastic Green-potential calculus and simultaneous covariance clock with averaged near-tip pole expansions to recover semimartingality and identify the quadratic variation and drift of the Loewner driver.

Source excerpts

“We first express h_t as harmonic measure plus a bounded Green potential. A height cutoff makes this potential a semimartingale up to swallowing; differences at two protected points then recover semimartingality of W. A common covariance clock records possible singular parts of both the driver and the fermion brackets. Finally, averaged expansions at the moving mark identify the cubic and quadratic poles, giving the variance, eliminating singular driver variation, and determining the drift.” · Show in paper

Stopped-Girsanov weak uniqueness for progressively measurable finite-energy drift functionals, localizing energy on finite capacity horizons without global Novikov conditions.

Source excerpts

“The proof uses attainment under the candidate solution to fix the stopping rule under the comparison Wiener law.” · Show in paper

Finite shielding exploration of monochromatic portals with a mandatory gate hierarchy and length-area uniformization to localize arbitrary Carath\'eodory approximations with diverging diameters.

Source excerpts

“We reduce the general case to that proposition by a finite edge exploration. The exploration will leave an ordinary bounded Dobrushin problem, with probability tending to one, and change the transferred curve by a vanishing amount.” · Show in paper