OctoInform Beta

OpenAI preprint, September 2026 · §4. The operator square · §4

Lemma 4.5 [Fixed squares]: Independence and essential total range of sides and adjoints make F*FF^*F and FF*FF^* of a measurable random operator rectangle almost surely fixed and injective

Fixed injective squares of random operator rectangles.

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

Four bounded operators between complex Hilbert spaces, F:H2→H1F\colon H_2\to H_1, G:H4→H3G\colon H_4\to H_3, K:H3→H1K\colon H_3\to H_1 and N:H4→H2N\colon H_4\to H_2, form a rectangle when they satisfy the first and the second rectangle identities FN=KG,F*K=NG*.FN = KG, \qquad F^*K = NG^*.(1) A random rectangle is a rectangle whose four sides are families of operators indexed by a probability space, F=(Fx)F = (F_x), G=(Gy)G = (G_y), K=(Kc)K = (K_c) and N=(Nd)N = (N_d), such that the identities hold for almost every choice of the parameters. Each side has a source square Fx*FxF_x^*F_x and a target square FxFx*F_xF_x^*, and the question is when each of them is almost surely equal to a fixed injective operator.

The library answers it for families with independent parameters, gives such families measurable polar parts, and relates the polar parts of the four sides of a rectangle. The rectangle it is built for, the grouped square defined below, has as sides the joining operators of conditional couplings between families of cells of a random array.

Operator Families

Let (Z,μ)(Z,\mu) be a measure space and F=(Fz)z∈ZF = (F_z)_{z\in Z} a family of bounded operators from a complex Hilbert space EE to a complex Hilbert space HH, with adjoint family F*=(Fz*)z∈ZF^* = (F_z^*)_{z\in Z}. A family F=(Fz)z∈ZF = (F_z)_{z\in Z} of bounded operators from EE to HH is strongly measurable if, for every v∈Ev\in E, the vector-valued map z↦Fzvz\mapsto F_zv is strongly measurable.✦ The essential kernel of FF is the subspace of the vectors v∈Ev\in E with Fzv=0F_zv = 0 for μ\mu-almost every zz✦, and the essential range ℛμ(F)\mathcal R_\mu(F) of FF is the orthogonal complement in HH of the essential kernel of F*F^*: ℛμ(F)={w∈H:Fz*w=0 for μ-almost every z}⟂.\mathcal R_\mu(F) = \{w\in H : F_z^*w = 0 \text{ for $\mu$-almost every } z\}^\perp.(2)✦ The family FF has essential total range if every closed subspace of HH that contains the range of FzF_z for μ\mu-almost every zz is HH itself.✦

A family FF has fixed injective squares with respect to μ\mu if there are injective bounded operators QEQ_E on EE and QHQ_H on HH such that Fz*Fz=QEandFzFz*=QHfor μ-almost every z.F_z^*F_z = Q_E \quad\text{and}\quad F_zF_z^* = Q_H \qquad \text{for $\mu$-almost every } z.(3)✦

The absolute value of a bounded operator V:E→HV\colon E\to H is the positive square root |V|=(V*V)1/2|V| = (V^*V)^{1/2}, an operator on EE, and a polar part of VV is a unitary operator U:E→HU\colon E\to H with U|V|=VU|V| = V.✦ For closed subspaces E0E_0 of EE and H0H_0 of HH, the compression of VV to E0E_0 and H0H_0 is the operator prH0V|E0\mathrm{pr}_{H_0}V|_{E_0} from E0E_0 to H0H_0, where prH0\mathrm{pr}_{H_0} is the orthogonal projection of HH onto H0H_0.✦

The rectangle spaces of four families F=(Fx)F = (F_x), G=(Gy)G = (G_y), K=(Kc)K = (K_c), N=(Nd)N = (N_d) with the domains and targets of (1), over measure spaces (ZF,μF)(Z_F,\mu_F), (ZG,μG)(Z_G,\mu_G), (ZK,μK)(Z_K,\mu_K) and (ZN,μN)(Z_N,\mu_N), are the closed subspaces D1=ℛ(F)∩ℛ(K)⊆H1,D2=ℛ(F*)∩ℛ(N)⊆H2,D3=ℛ(G)∩ℛ(K*)⊆H3,D4=ℛ(G*)∩ℛ(N*)⊆H4,\begin{aligned} D_1 &= \mathcal R(F)\cap\mathcal R(K)\subseteq H_1, & D_2 &= \mathcal R(F^*)\cap\mathcal R(N)\subseteq H_2,\\ D_3 &= \mathcal R(G)\cap\mathcal R(K^*)\subseteq H_3, & D_4 &= \mathcal R(G^*)\cap\mathcal R(N^*)\subseteq H_4, \end{aligned}(4) each essential range taken with respect to the law of its family’s parameter. The restricted families of the four families FF, GG, KK, NN are the compressions Fˆx:D2→D1\hat F_x\colon D_2\to D_1, Gˆy:D4→D3\hat G_y\colon D_4\to D_3, Kˆc:D3→D1\hat K_c\colon D_3\to D_1 and Nˆd:D4→D2\hat N_d\colon D_4\to D_2 of FxF_x, GyG_y, KcK_c and NdN_d to the rectangle spaces.

The Grouped Square

Let TT be a measurable automorphism of a standard Borel space XX and II a finite index set. An array of order nn over a space YY is a function from the index tuples InI^n to YY, a point of YInY^{I^n}✦; fixing one axis of an array of order n+1n+1 at a slot of II gives a plane of it, an array of order nn.✦ Initial array laws mXm^X are probability measures mnXm^X_n on XInX^{I^n}, one for each order nn, that are invariant under TT applied entrywise, consistent, in that every plane of an array of law mn+1Xm^X_{n+1} has law mnXm^X_n, and proper, in that for every proper subset JJ of II the planes fixing the last axis at the slots of JJ are independent.✦

An array object MM over mXm^X is a standard Borel space YY of states with an automorphism TYT_Y, an equivariant measurable map to XX and an equivariant measurable parameter map π\pi to a standard Borel space UU with an automorphism, together with consistent and proper probability measures mnm_n on YInY^{I^n}, the laws of MM, which are invariant under TYT_Y applied entrywise and carried to mnXm^X_n by the map to XX.✦ Two further conditions tie the laws to the parameters: given the parameter array π(y)\pi(y) of an array yy drawn from mnm_n, obtained by applying π\pi entrywise, the planes fixing the last axis at the slots of a proper subset of II remain independent, each with the conditional law of an array given its own parameter array; and at order zero the map to XX and π\pi are independent. For each order nn, κn(u)\kappa_n(u) denotes the conditional law under mnm_n of an array yy given π(y)=u\pi(y) = u.✦

An extension of MM is an array object M′M' over mXm^X with measurable maps from its states and its parameters to those of MM that commute with the automorphisms, the maps to XX and the parameter maps, and whose state map, applied entrywise, carries each law mn′m'_n to mnm_n.✦ The pullback of an extension M′M' of MM at order nn is the linear isometry from L2(mn)L^2(m_n) to L2(mn′)L^2(m'_n) given by composition with the entrywise state map of M′M'.✦ At order n+1n+1 an array object carries the data σ\sigma-algebras on YIn+1Y^{I^{n+1}}: the parameter σ\sigma-algebra 𝒫n+1\mathcal P_{n+1} and, for each j∈Ij\in I, the outside σ\sigma-algebra 𝒪n+1,j\mathcal O_{n+1,j} and the enlarged σ\sigma-algebra 𝒢n+1,j\mathcal G_{n+1,j}. An array object MM is saturated if, for every extension M′M' of it, every order nn, every data σ\sigma-algebra of order n+1n+1 and every f∈L2(mn+1)f\in L^2(m_{n+1}), the conditional expectation onto the data σ\sigma-algebra of M′M' of the pullback of ff is the pullback of the conditional expectation of ff onto the data σ\sigma-algebra of MM.✦

In an array yy of order n+2n+2 the first axis indexes rows and the last axis columns. The row ya,⋅y_{a,\cdot} and the column y⋅,jy_{\cdot,j} are the planes fixing the first axis at aa and the last axis at jj, arrays of order n+1n+1; the cell ya,jy_{a,j} is the array of order nn obtained by fixing both✦, and for sets AA of rows and SS of columns the cells of yy in A×SA\times S form the family yA×S=(ya,j)a∈A,j∈Sy_{A\times S} = (y_{a,j})_{a\in A,\, j\in S}.✦ For a parameter array ee of order n+2n+2 over UU, the law of cells meA×Sm_e^{A\times S} is the law of yA×Sy_{A\times S} when the rows ya,⋅y_{a,\cdot}, a∈Aa\in A, are drawn independently from κn+1(ea,⋅)\kappa_{n+1}(e_{a,\cdot}).✦

The joining operator of a coupling of two random points, with laws ν\nu and ρ\rho, is the operator from L2(ρ)L^2(\rho) to L2(ν)L^2(\nu) that sends a function of the second point to its conditional expectation given the first.✦ For a set AA of rows and sets RR, SS, S′S' of columns, the row joining operator at ee and at cells ξ\xi in A×RA\times R is the joining operator from L2(meA×S′)L^2(m_e^{A\times S'}) to L2(meA×S)L^2(m_e^{A\times S}) of the conditional law, under mn+2m_{n+2}, of the pair (yA×S,yA×S′)(y_{A\times S}, y_{A\times S'}) given π(y)=e\pi(y) = e and yA×R=ξy_{A\times R} = \xi.✦ For sets PP, AA, BB of rows and a set SS of columns, the column joining operator at ee and at cells ζ\zeta in P×SP\times S is the joining operator from L2(meB×S)L^2(m_e^{B\times S}) to L2(meA×S)L^2(m_e^{A\times S}) of the law of the pair (yA×S,yB×S)(y_{A\times S}, y_{B\times S}) when the columns y⋅,jy_{\cdot,j}, j∈Sj\in S, are drawn independently, each from κn+1(e⋅,j)\kappa_{n+1}(e_{\cdot,j}) conditioned on its cells in the rows of PP being those of ζ\zeta.✦

A triple partition of II is a partition I=P⊔A⊔BI = P\sqcup A\sqcup B into three nonempty disjoint sets.✦ Fix triple partitions (P,A,B)(P,A,B) of the rows and (P′,A′,B′)(P',A',B') of the columns, an order nn and a parameter array ee of order n+2n+2. The grouped square at ee consists of four families of operators between the spaces H1=L2(meA×A′),H2=L2(meA×B′),H3=L2(meB×A′),H4=L2(meB×B′).H_1 = L^2(m_e^{A\times A'}),\quad H_2 = L^2(m_e^{A\times B'}),\quad H_3 = L^2(m_e^{B\times A'}),\quad H_4 = L^2(m_e^{B\times B'}).(5) The sides FξeF^e_\xi and GηeG^e_\eta of the grouped square at ee are the row joining operators of the rows in AA and in BB from the columns B′B' to the columns A′A', at cells ξ\xi in A×P′A\times P' and η\eta in B×P′B\times P'; its sides KζeK^e_\zeta and NθeN^e_\theta are the column joining operators from the rows BB to the rows AA, in the columns A′A' and B′B', at cells ζ\zeta in P×A′P\times A' and θ\theta in P×B′P\times B'. The parameters ξ\xi, η\eta, ζ\zeta, θ\theta have the laws meA×P′m_e^{A\times P'}, meB×P′m_e^{B\times P'}, meP×A′m_e^{P\times A'} and meP×B′m_e^{P\times B'}, and the restricted families of the grouped square are written Fˆe\hat F^e, Gˆe\hat G^e, Kˆe\hat K^e and Nˆe\hat N^e.

The Claim

Theorem 1 (Fixed squares). If bounded operators FF, GG, KK, NN satisfy the rectangle identities (1), then KK*F=FNN*,K*KG=GN*N,F*FN=NG*G,FF*K=KGG*.KK^*F = FNN^*,\qquad K^*KG = GN^*N,\qquad F^*FN = NG^*G,\qquad FF^*K = KGG^*.(6)✦

Let FF, GG, KK, NN be families of bounded operators Fx:H2→H1F_x\colon H_2\to H_1, Gy:H4→H3G_y\colon H_4\to H_3, Kc:H3→H1K_c\colon H_3\to H_1 and Nd:H4→H2N_d\colon H_4\to H_2 over probability spaces (ZF,μF)(Z_F,\mu_F), (ZG,μG)(Z_G,\mu_G), (ZK,μK)(Z_K,\mu_K) and (ZN,μN)(Z_N,\mu_N), such that each of the eight families FF, GG, KK, NN, F*F^*, G*G^*, K*K^*, N*N^* has essential total range. If each identity of (6), with FxF_x, GyG_y, KcK_c, NdN_d in place of FF, GG, KK, NN, holds for almost every triple of the parameters it involves under the product of their laws, then F,G,K and N have fixed injective squares with respect to μF,μG,μK and μN.F,\ G,\ K \text{ and } N \text{ have fixed injective squares with respect to } \mu_F,\ \mu_G,\ \mu_K \text{ and } \mu_N.(7)✦ The same holds when H1H_1, H2H_2, H3H_3, H4H_4 are separable, the four families are strongly measurable and there are random parameters xx, yy, cc, dd on a probability space (Ω,μ)(\Omega,\mu), with laws μF\mu_F, μG\mu_G, μK\mu_K, μN\mu_N and any three of them independent, such that almost surely FxNd=KcGyF_xN_d = K_cG_y and Fx*Kc=NdGy*F_x^*K_c = N_dG_y^*: F,G,K and N have fixed injective squares with respect to μF,μG,μK and μN.F,\ G,\ K \text{ and } N \text{ have fixed injective squares with respect to } \mu_F,\ \mu_G,\ \mu_K \text{ and } \mu_N.(8)✦

Let (Fz)z∈Z(F_z)_{z\in Z} be a strongly measurable family of bounded operators from EE to HH, both separable, over a probability space (Z,μ)(Z,\mu), with fixed injective squares. Then there is a measurable family of unitary operators Uz:E→HU_z\colon E\to H such that, for μ\mu-almost every zz, the operator FzF_z is injective with dense range and Uz|Fz|=Fz.U_z|F_z| = F_z.(9)✦

If bounded operators FF, GG, KK, NN satisfy the rectangle identities (1), GG and NN are injective, and UFU_F, UGU_G, UKU_K, UNU_N are polar parts of FF, GG, KK, NN, then UFUN=UKUG.U_FU_N = U_KU_G.(10)✦

Let MM be a saturated array object, (P,A,B)(P,A,B) and (P′,A′,B′)(P',A',B') triple partitions of II, and n≥0n\ge 0. Then, for almost every parameter array ee of order n+2n+2 under the law of π(y)\pi(y) for yy drawn from mn+2m_{n+2}, Fˆe,Gˆe,Kˆe and Nˆe have fixed injective squares.\hat F^e,\ \hat G^e,\ \hat K^e \text{ and } \hat N^e \text{ have fixed injective squares.}(11)✦

In the paper, the result reads: for the four restricted families of the grouped square, each source square and each target square is almost surely independent of its parameter and injective on its space DiD_i, and the four polar parts are measurable families of unitaries between the spaces D1,…,D4D_1, \dots, D_4 that satisfy the first rectangle identity.

What Is Checked

The claim is stated over bounded operators between complex Hilbert spaces with their adjoints, absolute values and polar parts, over families of them indexed by probability spaces with the essential ranges, essential total range, strong measurability, fixed injective squares, rectangle spaces and compressions defined above, and, for the grouped square, over saturated array objects, triple partitions, laws of cells and the row and column joining operators built from the conditional laws of a random array.

Of these constructions the library proves that the state map of an extension, applied entrywise, is measure preserving from each law of the extension to the law of the same order of the array object.✦