Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 4 — consistency of the ideal HAC estimator Ω~T\widetilde\Omega_TΩT​

Proved
TPG.lemma_4_ideal_hac_consistency

by Wenqian Xing · Oct 12, 2026 · Mathlib 0df444a (Lean v4.33.1)

causal-inferencemarkov-chainnonstationaritypolicy-gradienttreatment-effect

Setting. SSS is a finite state space and the action Zt∈{0,1}Z_t\in\{0,1\}Zt​∈{0,1} is control or treatment. Time is t=0,1,…,T−1t=0,1,\dots,T-1t=0,1,…,T−1, which is the source's time t+1t+1t+1. Under action zzz at time ttt the state moves according to a transition matrix PztP_z^tPzt​, i.e. P(Xt+1=x′∣Xt=x,Zt=z)=Pzt(x,x′)\mathbb P(X_{t+1}=x'\mid X_t=x,Z_t=z)=P_z^t(x,x')P(Xt+1​=x′∣Xt​=x,Zt​=z)=Pzt​(x,x′); the matrices may depend on ttt arbitrarily. The initial state X0X_0X0​ has distribution ρ\rhoρ, and the reward YtY_tYt​ has distribution R(⋅∣Xt,Zt)R(\cdot\mid X_t,Z_t)R(⋅∣Xt​,Zt​) on R\mathbb RR, with R([−M,M]∣x,z)=1R([-M,M]\mid x,z)=1R([−M,M]∣x,z)=1 for all x,zx,zx,z and conditional independence from everything else given (Xt,Zt)(X_t,Z_t)(Xt​,Zt​). Under the policy πθ\pi_\thetaπθ​ every action is treatment with probability θ\thetaθ, independently of everything else; Eθ\mathbb E_\thetaEθ​, Pθ\mathbb P_\thetaPθ​ and Var⁡θ\operatorname{Var}_\thetaVarθ​ refer to the law of the trajectory (Xt,Zt,Yt)t<T(X_t,Z_t,Y_t)_{t<T}(Xt​,Zt​,Yt​)t<T​ under πθ\pi_\thetaπθ​.

Asymptotic setting. The truncation size k≥0k\ge0k≥0 is fixed and the horizon T→∞T\to\inftyT→∞; for each TTT the trajectory (Xt,Zt,Yt)t<T(X_t,Z_t,Y_t)_{t<T}(Xt​,Zt​,Yt​)t<T​ is generated under the experimental policy π1/2\pi_{1/2}π1/2​. Assumption 1 holds for all times with one constant γ∈(0,1)\gamma\in(0,1)γ∈(0,1). Write wtIPW=2(2Zt−1)w_t^{\mathrm{IPW}}=2(2Z_t-1)wtIPW​=2(2Zt​−1) and

Bt=∑i=max⁡(0,t−k)twiIPW Yt,μt=E[Bt],Bˉ=1T∑t<TBt=τ^k.B_t=\sum_{i=\max(0,t-k)}^{t}w_i^{\mathrm{IPW}}\,Y_t,\qquad \mu_t=\mathbb E[B_t],\qquad \bar B=\frac1T\sum_{t<T}B_t=\widehat\tau_k .Bt​=i=max(0,t−k)∑t​wiIPW​Yt​,μt​=E[Bt​],Bˉ=T1​t<T∑​Bt​=τk​.

Assumption 2 (asymptotic variance): 1TVar⁡(∑t<TBt)→σk2\frac1T\operatorname{Var}\bigl(\sum_{t<T}B_t\bigr)\to\sigma_k^2T1​Var(∑t<T​Bt​)→σk2​ as T→∞T\to\inftyT→∞, with 0<σk2<∞0<\sigma_k^2<\infty0<σk2​<∞.

The bandwidth LTL_TLT​ is a deterministic integer sequence with 0≤LT<T0\le L_T<T0≤LT​<T for every T≥1T\ge1T≥1, LT→∞L_T\to\inftyLT​→∞ and LT=O(T1/3)L_T=O(T^{1/3})LT​=O(T1/3), and wℓNW=1−ℓ/(LT+1)w_\ell^{\mathrm{NW}}=1-\ell/(L_T+1)wℓNW​=1−ℓ/(LT​+1).

The ideal HAC estimator centers at the time-specific means: with Vt=Bt−E[Bt]V_t=B_t-\mathbb E[B_t]Vt​=Bt​−E[Bt​],

Γ~ℓ=1T∑t+ℓ<TVt Vt+ℓ,Ω~T=Γ~0+2∑ℓ=1LTwℓNW Γ~ℓ.\widetilde\Gamma_\ell=\frac1T\sum_{t+\ell<T}V_t\,V_{t+\ell},\qquad \widetilde\Omega_T=\widetilde\Gamma_0+2\sum_{\ell=1}^{L_T}w_\ell^{\mathrm{NW}}\,\widetilde\Gamma_\ell .Γℓ​=T1​t+ℓ<T∑​Vt​Vt+ℓ​,ΩT​=Γ0​+2ℓ=1∑LT​​wℓNW​Γℓ​.

Lemma. Under Assumptions 1 and 2, Ω~T→pσk2\widetilde\Omega_T\xrightarrow{p}\sigma_k^2ΩT​p​σk2​ as T→∞T\to\inftyT→∞.

The ideal estimator is not computable from data, since the means E[Bt]\mathbb E[B_t]E[Bt​] are unknown; the lemma is the first step of the consistency proof for the feasible estimator of Theorem 2.

Formalization Note Convergence in probability is stated as P(∣Ω~T−σk2∣≥ε)→0\mathbb P(|\widetilde\Omega_T-\sigma_k^2|\ge\varepsilon)\to0P(∣ΩT​−σk2​∣≥ε)→0 for every ε>0\varepsilon>0ε>0, with the probability taken under the horizon-TTT law. The limit σk2\sigma_k^2σk2​ is a parameter vvv constrained by Assumption 2.

Preamble
import Mathlib
import Definitions.Def_TPG_Model
import Definitions.Def_TPG_Estimators
import Definitions.Def_TPG_Inference

open MeasureTheory ProbabilityTheory
open scoped BigOperators ENNReal
Formal statement
namespace TPG
theorem lemma_4_ideal_hac_consistency {S : Type} [Fintype S] [MeasurableSpace S] [MeasurableSingletonClass S]
    (E : Env S) (k : ℕ) (γ M v : ℝ) (L : ℕ → ℕ)
    (hγ0 : 0 < γ) (hγ1 : γ < 1) (hmix : ∀ T : ℕ, E.Mixing T γ) (hM : E.RewardBound M)
    (hvar : E.AsymptoticVariance k v) (hL : IsBandwidth L) :
    E.TendstoInProb (fun (T : ℕ) (ω : Traj S T) => E.hacIdeal T k (L T) ω) v := by sorry
end TPG
Source
Johari, Peng, Xing, Estimation of Treatment Effects Under Nonstationarity via the Truncated Policy Gradient Estimator, revised manuscript of October 2026 (revision of arXiv:2506.05308v3, https://arxiv.org/abs/2506.05308); numbering follows the revised manuscript. Appendix F.2, p. 42, Lemma 4; proof in Appendix H.2, pp. 53–54.
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

Read-back of the theorem TPG.lemma_4_ideal_hac_consistency.

1. Data quantified over. The statement is asserted for every choice of the following objects.

  • A finite set SSS, carrying a σ\sigmaσ-algebra in which every singleton is measurable. (As SSS is finite, this forces every subset of SSS to be measurable. SSS is not assumed to be non-empty.)

  • An environment EEE on SSS, which is a bundle of three things:

    • for every time t∈N={0,1,2,… }t\in\mathbb{N}=\lbrace 0,1,2,\dots\rbracet∈N={0,1,2,…} and every z∈{0,1}z\in\lbrace 0,1\rbracez∈{0,1} (a Boolean; 111 = true, 000 = false), a real S×SS\times SS×S matrix PtzP_t^{z}Ptz​ all of whose entries are ≥0\ge 0≥0 and each of whose rows sums to 111;
    • a function ρ:S→R\rho:S\to\mathbb{R}ρ:S→R with ρ(x)≥0\rho(x)\ge 0ρ(x)≥0 for every xxx and ∑x∈Sρ(x)=1\sum_{x\in S}\rho(x)=1∑x∈S​ρ(x)=1;
    • for every x∈Sx\in Sx∈S and z∈{0,1}z\in\lbrace 0,1\rbracez∈{0,1}, a Borel probability measure Rx,zR_{x,z}Rx,z​ on R\mathbb{R}R.

    The matrices PtzP_t^{z}Ptz​ are indexed by all t∈Nt\in\mathbb{N}t∈N and do not depend on any horizon. If SSS is empty, no environment exists (the requirement ∑xρ(x)=1\sum_x\rho(x)=1∑x​ρ(x)=1 cannot hold), so the statement is vacuous for empty SSS.

  • A natural number kkk (the value k=0k=0k=0 is allowed), three real numbers γ\gammaγ, MMM, vvv, and a function L:N→NL:\mathbb{N}\to\mathbb{N}L:N→N.

2. The probability spaces (a different one for each TTT). For each T∈NT\in\mathbb{N}T∈N let

ΩT=(S×{0,1}×R){0,1,…,T−1}\Omega_T=\bigl(S\times\lbrace 0,1\rbrace\times\mathbb{R}\bigr)^{\lbrace 0,1,\dots,T-1\rbrace}ΩT​=(S×{0,1}×R){0,1,…,T−1}

be the set of trajectories of length TTT, with the product σ\sigmaσ-algebra (the given σ\sigmaσ-algebra on SSS, all subsets of {0,1}\lbrace 0,1\rbrace{0,1}, the Borel σ\sigmaσ-algebra on R\mathbb{R}R). For ω∈ΩT\omega\in\Omega_Tω∈ΩT​ and t∈{0,…,T−1}t\in\lbrace 0,\dots,T-1\rbracet∈{0,…,T−1} write ωt=(Xt(ω),Zt(ω),Yt(ω))\omega_t=(X_t(\omega),Z_t(\omega),Y_t(\omega))ωt​=(Xt​(ω),Zt​(ω),Yt​(ω)) for the three coordinates at time ttt.

The measure PT\mathbb{P}_TPT​ on ΩT\Omega_TΩT​ is the finite mixture

PT=∑x∈ST, z∈{0,1}Tmax⁡(wT(x,z),0) ⨂t=0T−1(δxt⊗δzt⊗Rxt,zt),\mathbb{P}_T=\sum_{x\in S^T,\ z\in\lbrace 0,1\rbrace^T}\max\bigl(w_T(x,z),0\bigr)\ \bigotimes_{t=0}^{T-1}\Bigl(\delta_{x_t}\otimes\delta_{z_t}\otimes R_{x_t,z_t}\Bigr),PT​=x∈ST, z∈{0,1}T∑​max(wT​(x,z),0) t=0⨂T−1​(δxt​​⊗δzt​​⊗Rxt​,zt​​),

with weights

wT(x,z)=ρ(x0)⋅∏s=0T−2Pszs(xs,xs+1)⋅∏t=0T−112.w_T(x,z)=\rho(x_0)\cdot\prod_{s=0}^{T-2}P_s^{z_s}(x_s,x_{s+1})\cdot\prod_{t=0}^{T-1}\tfrac12 .wT​(x,z)=ρ(x0​)⋅s=0∏T−2​Pszs​​(xs​,xs+1​)⋅t=0∏T−1​21​.

Here δ\deltaδ denotes a point mass, Pszs(xs,xs+1)P_s^{z_s}(x_s,x_{s+1})Pszs​​(xs​,xs+1​) is the entry of the matrix PszsP_s^{z_s}Pszs​​ in row xsx_sxs​ and column xs+1x_{s+1}xs+1​, and in the last product the factor attached to time ttt is 12\tfrac1221​ if zt=1z_t=1zt​=1 and 1−121-\tfrac121−21​ if zt=0z_t=0zt​=0 (the underlying construction allows a per-time probability qtq_tqt​ in place of 12\tfrac1221​; in this theorem it is the constant 12\tfrac1221​). With the constant 12\tfrac1221​ every weight is non-negative, so the clipping max⁡(⋅,0)\max(\cdot,0)max(⋅,0) changes nothing, and the weights sum to 111, so PT\mathbb{P}_TPT​ is a probability measure.

In words, under PT\mathbb{P}_TPT​: Z0,…,ZT−1Z_0,\dots,Z_{T-1}Z0​,…,ZT−1​ are independent fair coin flips; X0X_0X0​ has distribution ρ\rhoρ; given the coin flips, X0,…,XT−1X_0,\dots,X_{T-1}X0​,…,XT−1​ is a time-inhomogeneous Markov chain in which the step from time sss to time s+1s+1s+1 uses the transition matrix PsZsP_s^{Z_s}PsZs​​ (rows are the current state, columns the next state); and, conditionally on the whole path (Xt,Zt)t<T(X_t,Z_t)_{t\lt T}(Xt​,Zt​)t<T​, the real numbers Y0,…,YT−1Y_0,\dots,Y_{T-1}Y0​,…,YT−1​ are independent with YtY_tYt​ distributed as RXt,ZtR_{X_t,Z_t}RXt​,Zt​​.

3. The statistics. Fix TTT and ω∈ΩT\omega\in\Omega_Tω∈ΩT​. Put a(1)=+2a(1)=+2a(1)=+2 and a(0)=−2a(0)=-2a(0)=−2 (in the code: a(z)=2 (2⋅1[z=1]−1)a(z)=2\,(2\cdot\mathbf{1}[z=1]-1)a(z)=2(2⋅1[z=1]−1)). For t∈{0,…,T−1}t\in\lbrace 0,\dots,T-1\rbracet∈{0,…,T−1} define

Bt(ω)=∑0≤i≤T−1i≤t≤i+ka(Zi(ω)) Yt(ω)=Yt(ω)∑i=max⁡(t−k, 0)ta(Zi(ω)),B_t(\omega)=\sum_{\substack{0\le i\le T-1\\ i\le t\le i+k}}a\bigl(Z_i(\omega)\bigr)\,Y_t(\omega)=Y_t(\omega)\sum_{i=\max(t-k,\,0)}^{t}a\bigl(Z_i(\omega)\bigr),Bt​(ω)=0≤i≤T−1i≤t≤i+k​∑​a(Zi​(ω))Yt​(ω)=Yt​(ω)i=max(t−k,0)∑t​a(Zi​(ω)),

and its mean under the length-TTT law,

mt(T)=∫ΩTBt dPT.m^{(T)}_t=\int_{\Omega_T}B_t\,d\mathbb{P}_T .mt(T)​=∫ΩT​​Bt​dPT​.

For a lag ℓ∈N\ell\in\mathbb{N}ℓ∈N define the mean-centred lag-ℓ\ellℓ product average

Γ~T,ℓ(ω)=1T∑0≤t,t′≤T−1t+ℓ=t′(Bt(ω)−mt(T))(Bt′(ω)−mt′(T))=1T∑t=0T−1−ℓ(Bt(ω)−mt(T))(Bt+ℓ(ω)−mt+ℓ(T)),\tilde\Gamma_{T,\ell}(\omega)=\frac1T\sum_{\substack{0\le t,\,t'\le T-1\\ t+\ell=t'}}\bigl(B_t(\omega)-m^{(T)}_t\bigr)\bigl(B_{t'}(\omega)-m^{(T)}_{t'}\bigr)=\frac1T\sum_{t=0}^{T-1-\ell}\bigl(B_t(\omega)-m^{(T)}_t\bigr)\bigl(B_{t+\ell}(\omega)-m^{(T)}_{t+\ell}\bigr),Γ~T,ℓ​(ω)=T1​0≤t,t′≤T−1t+ℓ=t′​∑​(Bt​(ω)−mt(T)​)(Bt′​(ω)−mt′(T)​)=T1​t=0∑T−1−ℓ​(Bt​(ω)−mt(T)​)(Bt+ℓ​(ω)−mt+ℓ(T)​),

and, for a truncation level Λ∈N\Lambda\in\mathbb{N}Λ∈N,

V~T,Λ(ω)=Γ~T,0(ω)+2∑ℓ=1Λ(1−ℓΛ+1)Γ~T,ℓ(ω).\tilde V_{T,\Lambda}(\omega)=\tilde\Gamma_{T,0}(\omega)+2\sum_{\ell=1}^{\Lambda}\Bigl(1-\frac{\ell}{\Lambda+1}\Bigr)\tilde\Gamma_{T,\ell}(\omega).V~T,Λ​(ω)=Γ~T,0​(ω)+2ℓ=1∑Λ​(1−Λ+1ℓ​)Γ~T,ℓ​(ω).

The random variable appearing in the conclusion is V~T:=V~T,L(T)\tilde V_T:=\tilde V_{T,L(T)}V~T​:=V~T,L(T)​, i.e. the truncation level is Λ=L(T)\Lambda=L(T)Λ=L(T). Note that the centring uses the exact means mt(T)m^{(T)}_tmt(T)​ (numbers determined by EEE, kkk, TTT), not sample averages.

4. Hypotheses. All of the following are assumed.

  1. 0<γ0\lt\gamma0<γ and γ<1\gamma\lt 1γ<1.

  2. (Contraction of every one-step kernel.) For every T∈NT\in\mathbb{N}T∈N, every t∈Nt\in\mathbb{N}t∈N with t<Tt\lt Tt<T, every z∈{0,1}z\in\lbrace 0,1\rbracez∈{0,1}, and every pair of functions f,f′:S→Rf,f':S\to\mathbb{R}f,f′:S→R that are probability vectors (all values ≥0\ge 0≥0, values summing to 111),

12∑x′∈S∣∑x∈Sf′(x) Ptz(x,x′)−∑x∈Sf(x) Ptz(x,x′)∣ ≤ γ⋅12∑x∈S∣f′(x)−f(x)∣.\frac12\sum_{x'\in S}\Bigl|\sum_{x\in S}f'(x)\,P_t^{z}(x,x')-\sum_{x\in S}f(x)\,P_t^{z}(x,x')\Bigr|\ \le\ \gamma\cdot\frac12\sum_{x\in S}\bigl|f'(x)-f(x)\bigr| .21​x′∈S∑​​x∈S∑​f′(x)Ptz​(x,x′)−x∈S∑​f(x)Ptz​(x,x′)​ ≤ γ⋅21​x∈S∑​​f′(x)−f(x)​.

Since TTT ranges over all natural numbers, this is the same as requiring the inequality for every t∈Nt\in\mathbb{N}t∈N (the instance T=0T=0T=0 is empty).

  1. (Reward bound.) For every x∈Sx\in Sx∈S and z∈{0,1}z\in\lbrace 0,1\rbracez∈{0,1}: Rx,z([−M,M])=1R_{x,z}\bigl([-M,M]\bigr)=1Rx,z​([−M,M])=1.

  2. (Asymptotic variance.) v>0v\gt 0v>0, and

lim⁡T→∞ 1T Var⁡PT(∑t=0T−1Bt)=v,\lim_{T\to\infty}\ \frac1T\,\operatorname{Var}_{\mathbb{P}_T}\Bigl(\sum_{t=0}^{T-1}B_t\Bigr)=v ,T→∞lim​ T1​VarPT​​(t=0∑T−1​Bt​)=v,

where for each TTT the variance is that of the function ω↦∑t=0T−1Bt(ω)\omega\mapsto\sum_{t=0}^{T-1}B_t(\omega)ω↦∑t=0T−1​Bt​(ω) on ΩT\Omega_TΩT​ under PT\mathbb{P}_TPT​.

  1. (Conditions on LLL.) All three of: (a) for every T∈NT\in\mathbb{N}T∈N with T>0T\gt 0T>0, L(T)<TL(T)\lt TL(T)<T; (b) L(T)→∞L(T)\to\inftyL(T)→∞ as T→∞T\to\inftyT→∞; (c) L(T)=O(T1/3)L(T)=O(T^{1/3})L(T)=O(T1/3) as T→∞T\to\inftyT→∞, meaning there are a real constant CCC and a T0T_0T0​ such that ∣L(T)∣≤C ∣T1/3∣|L(T)|\le C\,|T^{1/3}|∣L(T)∣≤C∣T1/3∣ for all T≥T0T\ge T_0T≥T0​ (with T1/3T^{1/3}T1/3 the real power).

5. Conclusion. For every real ε>0\varepsilon\gt 0ε>0,

lim⁡T→∞ PT({ω∈ΩT: ∣V~T,L(T)(ω)−v∣≥ε})=0,\lim_{T\to\infty}\ \mathbb{P}_T\Bigl(\bigl\lbrace\omega\in\Omega_T:\ \bigl|\tilde V_{T,L(T)}(\omega)-v\bigr|\ge\varepsilon\bigr\rbrace\Bigr)=0 ,T→∞lim​ PT​({ω∈ΩT​: ​V~T,L(T)​(ω)−v​≥ε})=0,

the limit being taken in [0,∞][0,\infty][0,∞]. That is, V~T,L(T)\tilde V_{T,L(T)}V~T,L(T)​ "converges in probability to vvv" in the sense that the TTT-th term is evaluated on the TTT-th probability space (ΩT,PT)(\Omega_T,\mathbb{P}_T)(ΩT​,PT​); there is no single probability space carrying all TTT at once.

6. Conventions and boundary cases contained in the literal statement.

  • The sample space and the measure change with the index TTT of the limit, both in hypothesis 4 and in the conclusion. The coin-flip probability is fixed at 12\tfrac1221​ throughout, and the weights a(z)a(z)a(z) are fixed at ±2\pm 2±2.
  • The event in the conclusion uses the non-strict inequality ≥ε\ge\varepsilon≥ε.
  • kkk is a fixed natural number (it does not vary with TTT); no relation between kkk and LLL or TTT is assumed. For k=0k=0k=0, Bt=a(Zt)YtB_t=a(Z_t)Y_tBt​=a(Zt​)Yt​. For t<kt\lt kt<k the window max⁡(t−k,0),…,t\max(t-k,0),\dots,tmax(t−k,0),…,t is cut off at time 000.
  • Each Γ~T,ℓ\tilde\Gamma_{T,\ell}Γ~T,ℓ​ is normalised by 1/T1/T1/T, not by the number T−ℓT-\ellT−ℓ of terms; for ℓ≥T\ell\ge Tℓ≥T the sum is empty and Γ~T,ℓ=0\tilde\Gamma_{T,\ell}=0Γ~T,ℓ​=0. By hypothesis 5(a) the lags used satisfy ℓ≤L(T)<T\ell\le L(T)\lt Tℓ≤L(T)<T when T>0T\gt 0T>0. If L(T)=0L(T)=0L(T)=0 (forced for T=1T=1T=1) the lag sum is empty and V~T,0=Γ~T,0\tilde V_{T,0}=\tilde\Gamma_{T,0}V~T,0​=Γ~T,0​.
  • T=0T=0T=0: Ω0\Omega_0Ω0​ is a single point with unit mass, 1/T1/T1/T is read as 000, all sums over ttt are empty, so every statistic equals 000; L(0)L(0)L(0) is unconstrained; the real power 01/30^{1/3}01/3 is 000. None of this affects limits as T→∞T\to\inftyT→∞.
  • The integral defining mt(T)m^{(T)}_tmt(T)​ is a Bochner integral, which by convention is 000 if the integrand is not integrable. The variance in hypothesis 4 is ∫(U−∫U dPT)2dPT\int\bigl(U-\int U\,d\mathbb{P}_T\bigr)^2d\mathbb{P}_T∫(U−∫UdPT​)2dPT​ computed as a [0,∞][0,\infty][0,∞]-valued integral and then converted to a real number with ∞↦0\infty\mapsto 0∞↦0. Under hypothesis 3 each YtY_tYt​ lies in [−M,M][-M,M][−M,M] PT\mathbb{P}_TPT​-almost surely, so ∣Bt∣≤2(k+1)M|B_t|\le 2(k+1)M∣Bt​∣≤2(k+1)M almost surely.
  • No sign condition is imposed on MMM. If M<0M\lt 0M<0 the interval [−M,M][-M,M][−M,M] is empty and hypothesis 3 cannot hold. If M=0M=0M=0 then every Yt=0Y_t=0Yt​=0 almost surely, the variance in hypothesis 4 is 000 for all TTT, and hypothesis 4 (v>0v\gt 0v>0) cannot hold. So the statement has content only for M>0M\gt 0M>0.
  • The existence of the limit vvv and its strict positivity are assumptions (hypothesis 4), not conclusions.
  • Hypothesis 2 is imposed separately on each single matrix PtzP_t^{z}Ptz​, for both values of zzz and every time ttt, with one common constant γ∈(0,1)\gamma\in(0,1)γ∈(0,1).
  • This theorem has no hypothesis on how the means mt(T)m^{(T)}_tmt(T)​ vary with ttt.
Human review
  • Endorsed by tianyipeng · Oct 12, 2026

  • Endorsed by Wenqian Xing · Oct 12, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me