OctoInformBeta

OpenAI preprint, September 2026 · §2. The finite-interval action inequality · §2

Theorem 2.1 [Action inequality]: For orthonormal amplitudes vanishing at interval endpoints, alongside upper-bound and penalized matrix-flow estimates

Action suprema over lower triangular matrices dominate integrated moments.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

Take a bounded open interval I=(l,r)I = (l, r) with l<rl < r, a natural number NN, a parameter σ=γ−1/2∈(0,1)\sigma = \gamma - 1/2 \in (0, 1) for a spectral exponent γ∈(1/2,3/2)\gamma \in (1/2, 3/2), and positive spectral parameters k1,…,kN>0k_1, \dots, k_N > 0. For an L2L^2-orthonormal family of NN real amplitudes u=(u1,…,uN)Tu = (u_1, \dots, u_N)^T on II vanishing at ll and rr, how does maximizing an action functional over linear combinations CuCu with real lower triangular N×NN \times N matrices CC control the integrated moments ∑i=1N∫−kiki(ki2−s2)σds\sum_{i=1}^N \int_{-k_i}^{k_i} (k_i^2 - s^2)^\sigma\,ds?

These notes go through the finite-interval action inequality and its companion matrix-flow estimates, together with the definitions they are built on.

The Definitions

For real endpoints ll and rr, the Sobolev space H01(l,r)H^1_0(l, r) consists of pairs (v,v′)(v, v') of real-valued functions on ℝ\mathbb{R} that are square-integrable with respect to Lebesgue measure on (l,r)(l, r), with vv continuous on [l,r][l, r], satisfying v(l)=v(r)=0v(l) = v(r) = 0, and obeying the weak derivative identity ∫(l,r)v(x)φ′(x)dx=−∫(l,r)v′(x)φ(x)dx\int_{(l, r)} v(x)\varphi'(x)\,dx = -\int_{(l, r)} v'(x)\varphi(x)\,dx(1)✦ for every smooth compactly supported test function φ:ℝ→ℝ\varphi : \mathbb{R} \to \mathbb{R} whose support lies in (l,r)(l, r).

For a natural number NN, the space ℒN\mathcal{L}_N of real lower triangular matrices is the set of real N×NN \times N matrices CC satisfying Cij=0C_{ij} = 0 for all indices ii and jj with i<ji < j.✦

Given N∈ℕN \in \mathbb{N}, real numbers σ,l,r\sigma, l, r, a vector k=(k1,…,kN)∈ℝNk = (k_1, \dots, k_N) \in \mathbb{R}^N with diagonal matrix K=diag⁡(k1,…,kN)K = \operatorname{diag}(k_1, \dots, k_N), a family u∈H01(l,r)Nu \in H^1_0(l, r)^N whose components are pairs (uj,uj′)∈H01(l,r)(u_j, u'_j) \in H^1_0(l, r), and a real N×NN \times N matrix CC, the finite-interval action ℰK(Cu)∈ℝ\mathcal{E}_K(Cu) \in \mathbb{R} is defined by ℰK(Cu)=∫(l,r)(∑i=1N(Cu′(x))i2+∑i=1Nki2(Cu(x))i2−(∑i=1N(Cu(x))i2)1+1/σ)dx,\mathcal{E}_K(Cu) = \int_{(l, r)} \left( \sum_{i=1}^N (Cu'(x))_i^2 + \sum_{i=1}^N k_i^2 (Cu(x))_i^2 - \left(\sum_{i=1}^N (Cu(x))_i^2\right)^{1 + 1/\sigma} \right) dx,(2)✦ where (Cu(x))i=∑j=1NCijuj(x)(Cu(x))_i = \sum_{j=1}^N C_{ij} u_j(x) and (Cu′(x))i=∑j=1NCijuj′(x)(Cu'(x))_i = \sum_{j=1}^N C_{ij} u'_j(x).

With N∈ℕN \in \mathbb{N}, σ∈ℝ\sigma \in \mathbb{R}, and vectors k,a,v∈ℝNk, a, v \in \mathbb{R}^N, the action density 𝒜σ,k(a,v)∈ℝ\mathcal{A}_{\sigma, k}(a, v) \in \mathbb{R} is 𝒜σ,k(a,v)=∑i=1Nvi2+∑i=1Nki2ai2−(∑i=1Nai2)1+1/σ.\mathcal{A}_{\sigma, k}(a, v) = \sum_{i=1}^N v_i^2 + \sum_{i=1}^N k_i^2 a_i^2 - \left(\sum_{i=1}^N a_i^2\right)^{1 + 1/\sigma}.(3)✦

For a natural number NN, the space SymN\mathrm{Sym}_N of real symmetric matrices is the set of real N×NN \times N matrices BB satisfying B=BTB = B^T.✦

To each N∈ℕN \in \mathbb{N}, vector k∈ℝNk \in \mathbb{R}^N, real N×NN \times N matrix BB, and scalar s∈ℝs \in \mathbb{R} corresponds the matrix pencil X(s)∈ℝN×NX(s) \in \mathbb{R}^{N \times N} defined by X(s)=diag⁡(k12,…,kN2)−2sB+s2IN,X(s) = \operatorname{diag}(k_1^2, \dots, k_N^2) - 2sB + s^2 I_N,(4)✦ where INI_N is the N×NN \times N identity matrix.

For σ,δ,y∈ℝ\sigma, \delta, y \in \mathbb{R}, the regularized scalar primitive Φσ,δ(y)∈ℝ\Phi_{\sigma, \delta}(y) \in \mathbb{R} is defined by Φσ,δ(y)=∫0y(max⁡(t,0)+δ)σ−3/2dt.\Phi_{\sigma, \delta}(y) = \int_0^y \bigl(\max(t, 0) + \delta\bigr)^{\sigma - 3/2}\,dt.(5)✦

The field normalization constant cσ∈ℝc_\sigma \in \mathbb{R} for σ∈ℝ\sigma \in \mathbb{R} is defined by cσ=σ∫−∞∞(1+s2)σ−3/2ds.c_\sigma = \frac{\sigma}{\int_{-\infty}^{\infty} (1 + s^2)^{\sigma - 3/2}\,ds}.(6)✦

Using (4), (5), and (6), the truncated matrix field MR(B)∈ℝN×NM_R(B) \in \mathbb{R}^{N \times N} for N∈ℕN \in \mathbb{N}, σ,δ,R∈ℝ\sigma, \delta, R \in \mathbb{R}, k∈ℝNk \in \mathbb{R}^N, and B∈ℝN×NB \in \mathbb{R}^{N \times N} is defined by MR(B)=cσ∫−RR(Φσ,δ(X(s))−Φσ,δ(s2)IN)ds,M_R(B) = c_\sigma \int_{-R}^{R} \Bigl(\Phi_{\sigma, \delta}(X(s)) - \Phi_{\sigma, \delta}(s^2) I_N\Bigr)\,ds,(7)✦ where Φσ,δ(X(s))\Phi_{\sigma, \delta}(X(s)) denotes Φσ,δ\Phi_{\sigma, \delta} applied to X(s)X(s) via continuous functional calculus.

In the limit R→∞R \to \infty, the regularized matrix field M(B)∈ℝN×NM(B) \in \mathbb{R}^{N \times N} for N∈ℕN \in \mathbb{N}, σ,δ∈ℝ\sigma, \delta \in \mathbb{R}, k∈ℝNk \in \mathbb{R}^N, and B∈ℝN×NB \in \mathbb{R}^{N \times N} is defined from (7) by M(B)=limR→∞MR(B).M(B) = \lim_{R \to \infty} M_R(B).(8)✦

When σ<1\sigma < 1 and δ>0\delta > 0, for any N∈ℕN \in \mathbb{N} and k∈ℝNk \in \mathbb{R}^N, the symmetric matrix field M:SymN→SymNM : \mathrm{Sym}_N \to \mathrm{Sym}_N is the restriction of the regularized matrix field (8) to SymN\mathrm{Sym}_N.✦

Associated with (8) for N∈ℕN \in \mathbb{N}, σ,δ∈ℝ\sigma, \delta \in \mathbb{R}, k∈ℝNk \in \mathbb{R}^N, and B∈ℝN×NB \in \mathbb{R}^{N \times N} is the matrix field primitive J(B)∈ℝJ(B) \in \mathbb{R}, defined by J(B)=−∫01tr⁡(BM(tB))dt.J(B) = -\int_0^1 \operatorname{tr}\bigl(B M(tB)\bigr)\,dt.(9)✦

On SymN×ℝN\mathrm{Sym}_N \times \mathbb{R}^N, for N∈ℕN \in \mathbb{N}, σ∈ℝ\sigma \in \mathbb{R}, and k∈ℝNk \in \mathbb{R}^N, the action defect Dact(B,a)∈ℝD_{\mathrm{act}}(B, a) \in \mathbb{R} is given by Dact(B,a)=(∑i=1Nai2)1+1/σ−∑i=1Nai((diag⁡(k12,…,kN2)−B2)a)i.D_{\mathrm{act}}(B, a) = \left(\sum_{i=1}^N a_i^2\right)^{1 + 1/\sigma} - \sum_{i=1}^N a_i \Bigl(\bigl(\operatorname{diag}(k_1^2, \dots, k_N^2) - B^2\bigr) a\Bigr)_i.(10)✦

The constraint defect Dcon(B,a)∈ℝD_{\mathrm{con}}(B, a) \in \mathbb{R} for N∈ℕN \in \mathbb{N}, σ,δ∈ℝ\sigma, \delta \in \mathbb{R}, k∈ℝNk \in \mathbb{R}^N, B∈SymNB \in \mathrm{Sym}_N, and a∈ℝNa \in \mathbb{R}^N is defined from (8) by Dcon(B,a)=∑i=1N∑j=1N(M(B)ij−aiaj)2.D_{\mathrm{con}}(B, a) = \sum_{i=1}^N \sum_{j=1}^N \bigl(M(B)_{ij} - a_i a_j\bigr)^2.(11)✦

The Claim

Theorem 1. Let N∈ℕN \in \mathbb{N}, let l<rl < r be real endpoints, let σ∈(0,1)\sigma \in (0, 1), and let k=(k1,…,kN)∈ℝNk = (k_1, \dots, k_N) \in \mathbb{R}^N satisfy ki>0k_i > 0 for every ii, with K=diag⁡(k)K = \operatorname{diag}(k). If u=(u1,…,uN)∈H01(l,r)Nu = (u_1, \dots, u_N) \in H^1_0(l, r)^N satisfies ∫(l,r)ui(x)uj(x)dx=1\int_{(l, r)} u_i(x) u_j(x)\,dx = 1 when i=ji = j and 00 when i≠ji \neq j for all indices ii and jj, then in the extended real numbers [−∞,∞][-\infty, \infty], ∑i=1N∫−kiki(ki2−s2)σds≤supC∈ℒNℰK(Cu).\sum_{i=1}^N \int_{-k_i}^{k_i} (k_i^2 - s^2)^\sigma\,ds \le \sup_{C \in \mathcal{L}_N} \mathcal{E}_K(Cu).(12)✦

In the paper, the result reads: let u=(u1,…,uN)T∈H01(I;ℝN)u = (u_1, \dots, u_N)^T \in H^1_0(I; \mathbb{R}^N) satisfy ∫IuuTdx=IN\int_I u u^T\,dx = I_N and let ki>0k_i > 0; then sup⁡C lower triangularℰK(Cu)≥∑i=1N∫−kiki(ki2−s2)σds\sup_{C \text{ lower triangular}} \mathcal{E}_K(Cu) \ge \sum_{i=1}^N \int_{-k_i}^{k_i} (k_i^2 - s^2)^\sigma\,ds, where the supremum is over all real lower triangular N×NN \times N matrices, I=(l,r)I = (l, r), and H01(I;ℝN)H^1_0(I; \mathbb{R}^N) stands for H01(l,r)NH^1_0(l, r)^N.

Theorem 2. Let N∈ℕN \in \mathbb{N}, let σ∈(0,1)\sigma \in (0, 1), let l≤rl \le r and SS be real numbers, and let k=(k1,…,kN)∈ℝNk = (k_1, \dots, k_N) \in \mathbb{R}^N satisfy ki>0k_i > 0 for every ii. Suppose u,v:ℝ→ℝNu, v : \mathbb{R} \to \mathbb{R}^N are continuous on [l,r][l, r] with u(l)=u(r)=0u(l) = u(r) = 0, each component uiu_i is differentiable on (l,r)(l, r) with derivative viv_i, and for all i,ji, j, ∫lrui(t)uj(t)dt\int_l^r u_i(t) u_j(t)\,dt equals 11 when i=ji = j and 00 when i≠ji \neq j. If ∫lr𝒜σ,k(Cu(t),Cv(t))dt≤S\int_l^r \mathcal{A}_{\sigma, k}(Cu(t), Cv(t))\,dt \le S for every C∈ℒNC \in \mathcal{L}_N, then ∑i=1N∫−kiki(ki2−s2)σds≤S.\sum_{i=1}^N \int_{-k_i}^{k_i} (k_i^2 - s^2)^\sigma\,ds \le S.(13)✦

Theorem 3. Let N∈ℕN \in \mathbb{N}, let σ∈(0,1)\sigma \in (0, 1), let δ>0\delta > 0, let l≤rl \le r and SS be real numbers, and let k=(k1,…,kN)∈ℝNk = (k_1, \dots, k_N) \in \mathbb{R}^N satisfy ki>0k_i > 0 for every ii, with K=diag⁡(k)K = \operatorname{diag}(k). Suppose u,v:ℝ→ℝNu, v : \mathbb{R} \to \mathbb{R}^N are continuous on [l,r][l, r] with u(l)=u(r)=0u(l) = u(r) = 0, each component uiu_i is differentiable on (l,r)(l, r) with derivative viv_i, and for all i,ji, j, ∫lrui(t)uj(t)dt\int_l^r u_i(t) u_j(t)\,dt equals 11 when i=ji = j and 00 when i≠ji \neq j. If ∫lr𝒜σ,k(Cu(t),Cv(t))dt≤S\int_l^r \mathcal{A}_{\sigma, k}(Cu(t), Cv(t))\,dt \le S for every C∈ℒNC \in \mathcal{L}_N, then J(−K)−J(K)≤S.J(-K) - J(K) \le S.(14)✦

Theorem 4. Let N∈ℕN \in \mathbb{N}, let σ∈(0,1)\sigma \in (0, 1), let δ,ε>0\delta, \varepsilon > 0, let l≤rl \le r and η\eta be real numbers, and let k∈ℝNk \in \mathbb{R}^N with K=diag⁡(k)K = \operatorname{diag}(k). Suppose a,v:ℝ→ℝNa, v : \mathbb{R} \to \mathbb{R}^N are continuous on [l,r][l, r] with a(l)=a(r)=0a(l) = a(r) = 0 and each component aia_i differentiable on (l,r)(l, r) with derivative viv_i, and suppose B:ℝ→SymNB : \mathbb{R} \to \mathrm{Sym}_N satisfies B(l)=KB(l) = K and B(r)=−KB(r) = -K, is differentiable within [l,r][l, r] at every t∈[l,r]t \in [l, r] with derivative B′(t)=ε−1(M(B(t))−a(t)a(t)T)B'(t) = \varepsilon^{-1}\bigl(M(B(t)) - a(t)a(t)^T\bigr), and satisfies Dact(B(t),a(t))−ε−1Dcon(B(t),a(t))≤ηD_{\mathrm{act}}(B(t), a(t)) - \varepsilon^{-1} D_{\mathrm{con}}(B(t), a(t)) \le \eta for every t∈[l,r]t \in [l, r]. Then J(−K)−J(K)≤∫lr𝒜σ,k(a(t),v(t))dt+(r−l)η.J(-K) - J(K) \le \int_l^r \mathcal{A}_{\sigma, k}(a(t), v(t))\,dt + (r - l)\eta.(15)✦

What Is Checked

The action inequality (12) is stated over N∈ℕN \in \mathbb{N}, l<rl < r, σ∈(0,1)\sigma \in (0, 1), positive vectors k∈ℝNk \in \mathbb{R}^N, orthonormal families in H01(l,r)NH^1_0(l, r)^N satisfying (1), lower triangular matrices in ℒN\mathcal{L}_N, and the finite-interval action (2); the upper-bound estimate (13) and the regularized bound (14) are stated over N∈ℕN \in \mathbb{N}, l≤rl \le r, σ∈(0,1)\sigma \in (0, 1), positive vectors k∈ℝNk \in \mathbb{R}^N, pairs (u,v)(u, v) continuous on [l,r][l, r] with u′=vu' = v on (l,r)(l, r) and u(l)=u(r)=0u(l) = u(r) = 0 whose components are orthonormal on [l,r][l, r], lower triangular matrices in ℒN\mathcal{L}_N, the action density (3), and, for (14), δ>0\delta > 0 and the matrix field primitive (9); and the penalized action inequality (15) is stated over N∈ℕN \in \mathbb{N}, l≤rl \le r, σ∈(0,1)\sigma \in (0, 1), δ,ε>0\delta, \varepsilon > 0, η∈ℝ\eta \in \mathbb{R}, k∈ℝNk \in \mathbb{R}^N, pairs (a,v)(a, v) continuous on [l,r][l, r] with a′=va' = v on (l,r)(l, r) and a(l)=a(r)=0a(l) = a(r) = 0, symmetric matrix flows in SymN\mathrm{Sym}_N with boundary values ±diag⁡(k)\pm\operatorname{diag}(k) and derivative within [l,r][l, r] driven by the symmetric matrix field and symmetric outer product, and the action and constraint defects (10) and (11).