Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

EPnI implies the full bosonic private-capacity conclusion

Open
QAlgorithms.Papers.Independent.Support50.P06.epni_privateCapacity

by Goku · Oct 23, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

Consider the full single-mode noiseless bosonic wiretap channel with vacuum environment, Bob transmissivity 0≤η≤10\le\eta\le10≤η≤1, and complementary Eve output from the same lossless beam splitter. Let Nˉ≥0\bar N\ge0Nˉ≥0 be a finite photon budget. Use unassisted classical private codes with arbitrary normalized mixed, possibly entangled block inputs, uniform messages, independent private discrete encoder randomness, and a normalized POVM decoder. The unconditional total input photon number, averaged over messages and randomness, is at most nNˉn\bar NnNˉ in an nnn-use block. For every message and random branch, decoding success exceeds 1−ϵ1-\epsilon1−ϵ; Holevo leakage about the message, after averaging the private randomness, tends to zero. Rates are ln⁡M/n\ln M/nlnM/n nats/use.

Assume global EPnI for all positive mode counts and transmissivities in [0,1][0,1][0,1], with independent normalized port states of finite input entropies; output entropy may be infinite. Then

Cp(η,Nˉ)={g(ηNˉ)−g((1−η)Nˉ),η>1/2,\0,η≤1/2,g(x)=(x+1)ln⁡(x+1)−xln⁡x.C_p(\eta,\bar N)=\begin{cases}g(\eta\bar N)-g((1-\eta)\bar N),&\eta>1/2,\0,&\eta\le1/2,\end{cases}\qquad g(x)=(x+1)\ln(x+1)-x\ln x.Cp​(η,Nˉ)={g(ηNˉ)−g((1−η)Nˉ),​η>1/2,\0,​η≤1/2,​g(x)=(x+1)ln(x+1)−xlnx.

This rate is asymptotically achievable with product coherent signal states and collective block decoding, and restricting the encoder to such states leaves the capacity unchanged. For every positive nnn, the unrestricted nnn-letter private-information supremum under total budget nNˉn\bar NnNˉ equals nCp(η,Nˉ)nC_p(\eta,\bar N)nCp​(η,Nˉ); grouping any positive number of parallel identical modes multiplies operational capacity by that number. On the η>1/2\eta>1/2η>1/2 branch, the single-use coherent ensemble with isotropic prior p(α)=e−∣α∣2/Nˉ/(πNˉ)p(\alpha)=e^{-|\alpha|^2/\bar N}/(\pi\bar N)p(α)=e−∣α∣2/Nˉ/(πNˉ) has private information g(ηNˉ)−g((1−η)Nˉ)g(\eta\bar N)-g((1-\eta)\bar N)g(ηNˉ)−g((1−η)Nˉ) and thermal average input of mean Nˉ\bar NNˉ. Individual coherent signals have pure outputs at both receivers. At Nˉ=0\bar N=0Nˉ=0 the prior is the point mass at vacuum.

This identifies the conditional operational capacity and its single-use signal ensemble, while allowing arbitrarily long codes.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P06
Formal statement
noncomputable section

namespace QAlgorithms.Papers.Independent.Support50.P06

open MeasureTheory Filter

open scoped ENNReal BigOperators Topology

open QAlgorithms.RealRegisters (Hilbert)

open QAlgorithms.Papers.Independent.Dependency50.P33 (State IsDensity rankOne)

open QAlgorithms.Papers.Independent.Next46.P32
  (NumberString FockKernel identicalThermalKernel vacuum productCoherentKernel
   jointMix bobMarginal charlieMarginal pureLoss broadcast mixedOutput
   spectralEntropyNats FockPOVM realFockTracePair modeEnergy complexNormal
   StrongConjecture2)

open QAlgorithms.Papers.Independent.Claimed50.P39 (g)

open QAlgorithms.Papers.Independent.Claimed50.P11 (thermalEntropyInverse)

attribute [local instance] Classical.propDecidable Classical.decEq

/-- PDF p.1 abstract and p.5 §VI: EPnI implies the entire conditional privacy
capacity conclusion, including additivity and Gaussian coherent achievability.
Source correction: (5) prints j=(m,k); secrecy here concerns m after averaging
k, as required by (6) and the proof. Nats use ln(M)/n; Nbar=0 uses delta_0.
This theorem does not assert the EPnI conjecture or an operational quantum code. -/
theorem epni_privateCapacity (hepni : EntropyPhotonNumberInequality)
    (eta Nbar : ℝ) (heta : 0 ≤ eta) (heta_one : eta ≤ 1) (hNbar : 0 ≤ Nbar) :
    holevoMessagePrivacyCapacityNats 1 eta Nbar =
      ENNReal.ofReal (if 1 / 2 < eta then g (eta * Nbar) - g ((1 - eta) * Nbar) else 0) ∧
    holevoMessagePrivacyCapacityNats 1 eta Nbar true = holevoMessagePrivacyCapacityNats 1 eta Nbar ∧
    privateAchievable 1 eta Nbar true
      (if 1 / 2 < eta then g (eta * Nbar) - g ((1 - eta) * Nbar) else 0) ∧
    (∀ n : ℕ, 0 < n → blockPrivateInformation n eta Nbar =
      (((n : ℝ) * (if 1 / 2 < eta then
        g (eta * Nbar) - g ((1 - eta) * Nbar) else 0) : ℝ) : EReal)) ∧
    (∀ width : ℕ, 0 < width → holevoMessagePrivacyCapacityNats width eta Nbar =
      (width : ℝ≥0∞) * holevoMessagePrivacyCapacityNats 1 eta Nbar) ∧
    (1 / 2 < eta →
      gaussianPrivateInformation eta Nbar = g (eta * Nbar) - g ((1 - eta) * Nbar) ∧
      coherentMixture Nbar = identicalThermalKernel 1 Nbar ∧
      ∀ alpha : ℂ,
        spectralEntropyNats (pureLoss 1 eta (productCoherentKernel 1 (fun _ => alpha))) = 0 ∧
        spectralEntropyNats ((charlieMarginal (broadcast 1 eta (productCoherentKernel 1 (fun _ => alpha))))) = 0) := by sorry

end QAlgorithms.Papers.Independent.Support50.P06

end
Source
Capacity of the Bosonic Wiretap Channel and the Entropy Photon-Number Inequality, arXiv:0801.0841v1, PDF p. 5, Abstract and §VI: EPnI implies the wiretap-capacity theorem
Read-back

What the Lean code literally says, in plain math · gpt-6.1-sol

Assume the entropy photon number inequality defined below, and fix real parameters 0≤η≤10\le\eta\le10≤η≤1 and Nˉ≥0\bar N\ge0Nˉ≥0. The kernels, entropies, codes, and optimizations occurring in the conclusion have the following literal meanings. For any positive integer ddd, let Nd={0,…,d−1}→N\mathcal N_d=\{0,\ldots,d-1\}\to\mathbb NNd​={0,…,d−1}→N. A density kernel XXX on this index set has a representation Xab=∑rlr(a)rr(b)‾X_{ab}=\sum_r l_r(a)\overline{r_r(b)}Xab​=∑r​lr​(a)rr​(b)​, where lr,rr∈ℓ2(Nd;C)l_r,r_r\in\ell^2(\mathcal N_d;\mathbb C)lr​,rr​∈ℓ2(Nd​;C) and ∑r∥lr∥∥rr∥\sum_r\|l_r\|\|r_r\|∑r​∥lr​∥∥rr​∥ is summable; every finite quadratic form ∑a,b∈Fza‾Xabzb\sum_{a,b\in F}\overline{z_a}X_{ab}z_b∑a,b∈F​za​​Xab​zb​ is real and nonnegative; and the real diagonal entries are summable with sum 111. Define S(X)∈[0,∞]S(X)\in[0,\infty]S(X)∈[0,∞] to be the infimum of ∑j[−λjlog⁡λj]+\sum_j[-\lambda_j\log\lambda_j]_+∑j​[−λj​logλj​]+​ over representations Xab=∑jλjvj(a)vj(b)‾X_{ab}=\sum_j\lambda_jv_j(a)\overline{v_j(b)}Xab​=∑j​λj​vj​(a)vj​(b)​ with nonnegative summable weights of total 111, unit norm vectors for positive weights, and mutually orthogonal vectors for distinct positive weights; zero-weight vectors are unrestricted. If there is no such representation, S(X)=∞S(X)=\inftyS(X)=∞. Here [r]+[r]_+[r]+​ is the nonnegative part of a real number embedded in [0,∞][0,\infty][0,∞]. Put

g(x)=(x+1)log⁡(x+1)−xlog⁡x,h(s)={0,s≤0,inf⁡{K≥0:s≤g(K)},s>0,Ed(s)={∞,s=∞,[h(s/d)]+,s<∞.g(x)=(x+1)\log(x+1)-x\log x,\qquad h(s)= \begin{cases} 0,&s\le0,\\ \inf\{K\ge0:s\le g(K)\},&s>0, \end{cases} \qquad E_d(s)= \begin{cases} \infty,&s=\infty,\\ [h(s/d)]_+,&s<\infty. \end{cases}g(x)=(x+1)log(x+1)−xlogx,h(s)={0,inf{K≥0:s≤g(K)},​s≤0,s>0,​Ed​(s)={∞,[h(s/d)]+​,​s=∞,s<∞.​

The logarithm is natural, with the total convention log⁡0=0\log0=0log0=0; an empty real infimum defining hhh is 000. For nonnegative integers m,l,p,qm,l,p,qm,l,p,q, define bt(m,l;p,q)=0b_t(m,l;p,q)=0bt​(m,l;p,q)=0 unless m+l=p+qm+l=p+qm+l=p+q, and otherwise define

bt(m,l;p,q)=m! l!p! q!∑0≤u≤pu≤m, m−u≤q(pu)(qm−u)(t)u(1−t) p−u+(m−u)(−t) q−(m−u).b_t(m,l;p,q)= \sqrt{\frac{m!\,l!}{p!\,q!}} \sum_{\substack{0\le u\le p\\u\le m,\ m-u\le q}} \binom pu\binom q{m-u} (\sqrt t)^u (\sqrt{1-t})^{\,p-u+(m-u)} (-\sqrt t)^{\,q-(m-u)}.bt​(m,l;p,q)=p!q!m!l!​​0≤u≤pu≤m, m−u≤q​∑​(up​)(m−uq​)(t​)u(1−t​)p−u+(m−u)(−t​)q−(m−u).

All natural-number subtraction is truncated at zero. Set Bd,t(u,a)=∏i=0d−1bt(u1(i),u2(i);a1(i),a2(i))B_{d,t}(u,a)=\prod_{i=0}^{d-1}b_t(u_1(i),u_2(i);a_1(i),a_2(i))Bd,t​(u,a)=∏i=0d−1​bt​(u1​(i),u2​(i);a1​(i),a2​(i)), and define

Jd,t(X,Y)uv=∑a∈Nd2∑b∈Nd2Bd,t(u,a)Xa1b1Ya2b2Bd,t(v,b)‾,Md,t(X,Y)ab=∑e∈NdJd,t(X,Y)(a,e),(b,e).J_{d,t}(X,Y)_{uv} =\sum_{a\in\mathcal N_d^2}\sum_{b\in\mathcal N_d^2} B_{d,t}(u,a)X_{a_1b_1}Y_{a_2b_2}\overline{B_{d,t}(v,b)}, \qquad M_{d,t}(X,Y)_{ab} =\sum_{e\in\mathcal N_d}J_{d,t}(X,Y)_{(a,e),(b,e)}.Jd,t​(X,Y)uv​=a∈Nd2​∑​b∈Nd2​∑​Bd,t​(u,a)Xa1​b1​​Ya2​b2​​Bd,t​(v,b)​,Md,t​(X,Y)ab​=e∈Nd​∑​Jd,t​(X,Y)(a,e),(b,e)​.

Let Vd(a,b)V_d(a,b)Vd​(a,b) be 111 when a=b=0a=b=\mathbf0a=b=0 and 000 otherwise. The broadcast kernel is Jd,t(X,Vd)J_{d,t}(X,V_d)Jd,t​(X,Vd​); its two marginals used here are

Ld,t(X)ab=∑eJd,t(X,Vd)(a,e),(b,e),Cd,t(X)ab=∑eJd,t(X,Vd)(e,a),(e,b).L_{d,t}(X)_{ab} =\sum_eJ_{d,t}(X,V_d)_{(a,e),(b,e)}, \qquad C_{d,t}(X)_{ab} =\sum_eJ_{d,t}(X,V_d)_{(e,a),(e,b)}.Ld,t​(X)ab​=e∑​Jd,t​(X,Vd​)(a,e),(b,e)​,Cd,t​(X)ab​=e∑​Jd,t​(X,Vd​)(e,a),(e,b)​.

All unrestricted complex or real series use total sums, taking value zero when nonsummable. The assumed entropy photon number inequality says that, for every positive integer ddd, every t∈[0,1]t\in[0,1]t∈[0,1], and every pair of density kernels X,YX,YX,Y with S(X),S(Y)<∞S(X),S(Y)<\inftyS(X),S(Y)<∞,

[t]+Ed(S(X))+[1−t]+Ed(S(Y))≤Ed(S(Md,t(X,Y))).[t]_+E_d(S(X))+[1-t]_+E_d(S(Y)) \le E_d(S(M_{d,t}(X,Y))).[t]+​Ed​(S(X))+[1−t]+​Ed​(S(Y))≤Ed​(S(Md,t​(X,Y))).

To specify the capacity, write

Ed(X)=∑i=0d−1∑a∈Nd[a(i)Re⁡Xaa]+,⟨X,F⟩tr,R=Re⁡∑a∑bXabFba.\mathcal E_d(X)= \sum_{i=0}^{d-1}\sum_{a\in\mathcal N_d} [a(i)\operatorname{Re}X_{aa}]_+, \qquad \langle X,F\rangle_{\mathrm{tr},\mathbb R} =\operatorname{Re}\sum_a\sum_bX_{ab}F_{ba}.Ed​(X)=i=0∑d−1​a∈Nd​∑​[a(i)ReXaa​]+​,⟨X,F⟩tr,R​=Rea∑​b∑​Xab​Fba​.

A private code of length ddd and message count MMM consists of d>0d>0d>0, M>0M>0M>0, a nonnegative summable real sequence (pk)k∈N(p_k)_{k\in\mathbb N}(pk​)k∈N​ of total 111, density kernels ρm,k\rho_{m,k}ρm,k​ for 0≤m<M0\le m<M0≤m<M and every k∈Nk\in\mathbb Nk∈N, and decoder kernels FmF_mFm​. Each FmF_mFm​ has real nonnegative quadratic forms on every finite set, and they satisfy the coefficientwise completeness condition

∑m=0M−1(Fm)ab={1,a=b,0,a≠b.\sum_{m=0}^{M-1}(F_m)_{ab}= \begin{cases}1,&a=b,\\0,&a\ne b.\end{cases}m=0∑M−1​(Fm​)ab​={1,0,​a=b,a=b.​

Its photon budget, Eve kernels, and leakage are

B=1M∑m=0M−1∑k=0∞[pk]+Ed(ρm,k),Zm(a,b)=∑k=0∞pk Cd,η(ρm,k)ab,Zˉ=1M∑m=0M−1Zm,L=S(Zˉ)−1M∑m=0M−1S(Zm).\begin{aligned} \mathcal B&=\frac1M\sum_{m=0}^{M-1}\sum_{k=0}^{\infty} [p_k]_+\mathcal E_d(\rho_{m,k}),\\ Z_m(a,b)&=\sum_{k=0}^{\infty}p_k\,C_{d,\eta}(\rho_{m,k})_{ab}, \qquad \bar Z=\frac1M\sum_{m=0}^{M-1}Z_m,\\ \mathcal L&=S(\bar Z)-\frac1M\sum_{m=0}^{M-1}S(Z_m). \end{aligned}BZm​(a,b)L​=M1​m=0∑M−1​k=0∑∞​[pk​]+​Ed​(ρm,k​),=k=0∑∞​pk​Cd,η​(ρm,k​)ab​,Zˉ=M1​m=0∑M−1​Zm​,=S(Zˉ)−M1​m=0∑M−1​S(Zm​).​

The budget and leakage use arithmetic in [0,∞][0,\infty][0,∞]; in particular the leakage subtraction is truncated subtraction, including ∞−∞=0\infty-\infty=0∞−∞=0. Decoding with parameter ϵ\epsilonϵ means the strict inequality

1−ϵ<⟨Ld,η(ρm,k),Fm⟩tr,Rfor every m,k,1-\epsilon< \langle L_{d,\eta}(\rho_{m,k}),F_m\rangle_{\mathrm{tr},\mathbb R} \quad\text{for every }m,k,1−ϵ<⟨Ld,η​(ρm,k​),Fm​⟩tr,R​for every m,k,

including indices kkk with pk=0p_k=0pk​=0. For amplitudes α=(αi)i<d\alpha=(\alpha_i)_{i<d}α=(αi​)i<d​, define

cα(a)=∏i=0d−1e−∣αi∣2/2αia(i)a(i)!,Pα(a,b)=cα(a)cα(b)‾.c_\alpha(a)= \prod_{i=0}^{d-1} \frac{e^{-|\alpha_i|^2/2}\alpha_i^{a(i)}}{\sqrt{a(i)!}}, \qquad P_\alpha(a,b)=c_\alpha(a)\overline{c_\alpha(b)}.cα​(a)=i=0∏d−1​a(i)!​e−∣αi​∣2/2αia(i)​​,Pα​(a,b)=cα​(a)cα​(b)​.

A coherent private code is one for which some amplitudes αm,k\alpha_{m,k}αm,k​ satisfy ρm,k=Pαm,k\rho_{m,k}=P_{\alpha_{m,k}}ρm,k​=Pαm,k​​ for every m,km,km,k. For a width w∈Nw\in\mathbb Nw∈N and a flag b∈{false,true}b\in\{\mathrm{false},\mathrm{true}\}b∈{false,true}, say Awb(R)A_w^b(R)Awb​(R) holds when R≥0R\ge0R≥0 and, for every 0<ϵ<10<\epsilon<10<ϵ<1 and every δ>0\delta>0δ>0, there exists n0>0n_0>0n0​>0 such that for every integer n≥n0n\ge n_0n≥n0​ there is such a private code of length nwnwnw, with some positive message count MMM, satisfying

B≤[nwNˉ]+,L≤[ϵ]+,R−δ≤log⁡Mn,\mathcal B\le[nw\bar N]_+,\qquad \mathcal L\le[\epsilon]_+,\qquad R-\delta\le\frac{\log M}{n},B≤[nwNˉ]+​,L≤[ϵ]+​,R−δ≤nlogM​,

and the decoding inequalities above, with the additional coherent-code requirement when b=trueb=\mathrm{true}b=true. Thus the rate denominator is nnn, whereas the physical code length is nwnwnw. Define

Cwb=sup⁡{[R]+:R∈R, Awb(R)}in [0,∞].\mathsf C_w^b=\sup\{[R]_+:R\in\mathbb R,\ A_w^b(R)\} \quad\text{in }[0,\infty].Cwb​=sup{[R]+​:R∈R, Awb​(R)}in [0,∞].

For the finite-ensemble quantity, let an ensemble on ddd modes consist of a positive integer LLL, nonnegative real weights qjq_jqj​, 0≤j<L0\le j<L0≤j<L, of total 111, and density kernels XjX_jXj​. Put Xˉ=∑jqjXj\bar X=\sum_jq_jX_jXˉ=∑j​qj​Xj​, and define

Hd,η(X)=S(Ld,η(X))real−S(Cd,η(X))real,H_{d,\eta}(X)=S(L_{d,\eta}(X))^{\mathrm{real}} -S(C_{d,\eta}(X))^{\mathrm{real}},Hd,η​(X)=S(Ld,η​(X))real−S(Cd,η​(X))real,

where sreal=ss^{\mathrm{real}}=ssreal=s for finite s∈[0,∞]s\in[0,\infty]s∈[0,∞], and ∞real=0\infty^{\mathrm{real}}=0∞real=0. Let

Pd=sup⁡{Hd,η(Xˉ)−∑j=0L−1qjHd,η(Xj):∑j[qj]+Ed(Xj)≤[dNˉ]+},\mathsf P_d= \sup\left\{ H_{d,\eta}(\bar X)-\sum_{j=0}^{L-1}q_jH_{d,\eta}(X_j): \sum_j[q_j]_+\mathcal E_d(X_j)\le[d\bar N]_+ \right\},Pd​=sup{Hd,η​(Xˉ)−j=0∑L−1​qj​Hd,η​(Xj​):j∑​[qj​]+​Ed​(Xj​)≤[dNˉ]+​},

where the supremum is in the extended real numbers and ranges over all such finite ensembles. Finally, for V≥0V\ge0V≥0, let γV\gamma_VγV​ be the point mass at 000 when V=0V=0V=0, and otherwise the measure on C\mathbb CC with density e−∣α∣2/V/(πV)e^{-|\alpha|^2/V}/(\pi V)e−∣α∣2/V/(πV) relative to planar volume. Define the one-mode kernel QVQ_VQV​ coefficientwise by QV(a,b)=∫Pα(a,b) dγV(α)Q_V(a,b)=\int P_\alpha(a,b)\,d\gamma_V(\alpha)QV​(a,b)=∫Pα​(a,b)dγV​(α), and define

Γη,V=H1,η(QV)−∫H1,η(Pα) dγV(α).\Gamma_{\eta,V} =H_{1,\eta}(Q_V)-\int H_{1,\eta}(P_\alpha)\,d\gamma_V(\alpha).Γη,V​=H1,η​(QV​)−∫H1,η​(Pα​)dγV​(α).

These integrals are the total integrals in the definitions; a nonintegrable integrand has integral zero. With all these meanings, put

F={g(ηNˉ)−g((1−η)Nˉ),η>12,0,η≤12.F= \begin{cases} g(\eta\bar N)-g((1-\eta)\bar N),&\eta>\tfrac12,\\ 0,&\eta\le\tfrac12. \end{cases}F={g(ηNˉ)−g((1−η)Nˉ),0,​η>21​,η≤21​.​

The declaration asserts simultaneously that

C1false=[F]+,C1true=C1false,A1true(F),\mathsf C_1^{\mathrm{false}}=[F]_+,\qquad \mathsf C_1^{\mathrm{true}}=\mathsf C_1^{\mathrm{false}},\qquad A_1^{\mathrm{true}}(F),C1false​=[F]+​,C1true​=C1false​,A1true​(F),

that Pd=dF\mathsf P_d=dFPd​=dF as an extended real number for every positive integer ddd, and that

Cwfalse=w C1falsefor every positive integer w.\mathsf C_w^{\mathrm{false}} =w\,\mathsf C_1^{\mathrm{false}} \quad\text{for every positive integer }w.Cwfalse​=wC1false​for every positive integer w.

In addition, if η>12\eta>\tfrac12η>21​, it asserts

Γη,Nˉ=g(ηNˉ)−g((1−η)Nˉ),QNˉ=T1,Nˉ,\Gamma_{\eta,\bar N}=g(\eta\bar N)-g((1-\eta)\bar N), \qquad Q_{\bar N}=T_{1,\bar N},Γη,Nˉ​=g(ηNˉ)−g((1−η)Nˉ),QNˉ​=T1,Nˉ​,

where T1,NˉT_{1,\bar N}T1,Nˉ​ is diagonal with

T1,Nˉ(a,a)=(Nˉ/(Nˉ+1))a(0)Nˉ+1,T_{1,\bar N}(a,a)=\frac{(\bar N/(\bar N+1))^{a(0)}}{\bar N+1},T1,Nˉ​(a,a)=Nˉ+1(Nˉ/(Nˉ+1))a(0)​,

and it asserts, for every α∈C\alpha\in\mathbb Cα∈C, that both S(L1,η(Pα))=0S(L_{1,\eta}(P_\alpha))=0S(L1,η​(Pα​))=0 and S(C1,η(Pα))=0S(C_{1,\eta}(P_\alpha))=0S(C1,η​(Pα​))=0. The parameter range includes Nˉ=0\bar N=0Nˉ=0, η=0,1\eta=0,1η=0,1, and η=12\eta=\tfrac12η=21​; the last of these uses the zero branch of FFF.

Human review
  • Endorsed by Shuze Chen · Oct 24, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 24, 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