OctoInformBeta

OpenAI preprint, September 2026 · §3. A unitary with a large fixed space

§3 (setting): incidence data and the model map: Horizontal directions and smooth homotopies for maps preserving unitary elements

Unitary coordinates for the incidence model.

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

Fix finite index sets KK and RR, and put k=|K|k=|K|, r=|R|r=|R|, and m=k+rm=k+r. The fixed space problem concerns a unitary matrix on ℂK⊔R\mathbb C^{K\sqcup R} together with a kk-plane that it fixes pointwise. For a nonnegative integer dd, write U(d)U(d) for the unitary group of a complex inner product space of dimension dd. Unitary coordinates retain a choice of Q∈U(m)Q\in U(m) representing that plane as Q(ℂK⊕0)Q(\mathbb C^K\oplus 0).

For a finite index set JJ, the algebra MJ(ℂ)M_J(\mathbb C) consists of complex square matrices whose rows and columns are indexed by JJ.✦ Write T*T^* for the conjugate transpose of TT. The real vector space 𝔲(J)\mathfrak u(J) consists of those Z∈MJ(ℂ)Z\in M_J(\mathbb C) satisfying Z*=−ZZ^*=-Z.

For X∈MK(ℂ)X\in M_K(\mathbb C) and Y∈MR(ℂ)Y\in M_R(\mathbb C), the block diagonal matrix diag⁡(X,Y)∈MK⊔R(ℂ)\operatorname{diag}(X,Y)\in M_{K\sqcup R}(\mathbb C) is defined by (1); IJI_J denotes the identity matrix indexed by JJ. diag⁡(X,Y)=(X00Y).\operatorname{diag}(X,Y)= \begin{pmatrix}X&0\\0&Y\end{pmatrix}.(1)✦ For a normed algebra BB with adjoint, write U(B)U(B) for its unitary group, and write Cℝ∞(B,C)C^\infty_{\mathbb R}(B,C) for the maps from BB to another such algebra CC that are smooth over ℝ\mathbb R.

The Definition

Definition 1. The triple matrix algebra 𝒜\mathcal A is the product algebra in (2). 𝒜=MK(ℂ)×MK⊔R(ℂ)×MR(ℂ).\mathcal A=M_K(\mathbb C)\times M_{K\sqcup R}(\mathbb C) \times M_R(\mathbb C).(2)✦ The horizontal tangent space ℋ\mathcal H is the real vector space in (3), where MR×K(ℂ)M_{R\times K}(\mathbb C) consists of complex matrices with rows indexed by RR and columns indexed by KK. ℋ=𝔲(K)×𝔲(R)×MR×K(ℂ).\mathcal H=\mathfrak u(K)\times\mathfrak u(R) \times M_{R\times K}(\mathbb C).(3)✦ Given a matrix family f:ℝ×MK(ℂ)→MK⊔R(ℂ)f:\mathbb R\times M_K(\mathbb C)\longrightarrow M_{K\sqcup R}(\mathbb C), its incidence map Φf:ℝ×𝒜→MK⊔R(ℂ)\Phi_f:\mathbb R\times\mathcal A\longrightarrow M_{K\sqcup R}(\mathbb C) is defined by (4), with (X,Q,V)∈𝒜(X,Q,V)\in\mathcal A. Φf(t;X,Q,V)=f(t,X)Qdiag⁡(IK,V*)Q*.\Phi_f(t;X,Q,V)=f(t,X)\,Q\operatorname{diag}(I_K,V^*)Q^*.(4)✦ For normed algebras BB and CC with adjoints, the class 𝒰(B,C)\mathcal U(B,C) of smooth unitary maps consists of the maps specified in (5). 𝒰(B,C)={F∈Cℝ∞(B,C):F(U(B))⊆U(C)}.\mathcal U(B,C)= \bigl\{F\in C^\infty_{\mathbb R}(B,C):F(U(B))\subseteq U(C)\bigr\}.(5)✦ The model unitary map μ∈𝒰(MK(ℂ),MK⊔R(ℂ))\mu\in\mathcal U(M_K(\mathbb C),M_{K\sqcup R}(\mathbb C)) is given by (6). μ(X)=diag⁡(X,−IR),X∈MK(ℂ).\mu(X)=\operatorname{diag}(X,-I_R), \qquad X\in M_K(\mathbb C).(6)✦ An elementary smooth homotopy from FF to GG in 𝒰(B,C)\mathcal U(B,C) is a smooth map H:ℝ×B→CH:\mathbb R\times B\longrightarrow C satisfying (7); write F⇝GF\rightsquigarrow G when such a map exists. H(t,u)∈U(C)(t∈ℝ,u∈U(B)),H(0,u)=F(u),H(1,u)=G(u)(u∈U(B)).\begin{gathered} H(t,u)\in U(C)\qquad(t\in\mathbb R,\ u\in U(B)),\\ H(0,u)=F(u),\qquad H(1,u)=G(u)\qquad(u\in U(B)). \end{gathered}(7)✦ The smooth homotopy relation ∼\sim on 𝒰(B,C)\mathcal U(B,C) is defined by the finite chains in (8), allowing a chain of length zero. F∼G⇔there exist q∈ℕ and F0,…,Fq∈𝒰(B,C)with F0=F,Fq=G, and, for every 0≤j<q,Fj⇝Fj+1orFj+1⇝Fj.\begin{split} F\sim G\quad\Longleftrightarrow\quad& \text{there exist }q\in\mathbb N\text{ and }F_0,\ldots,F_q\in\mathcal U(B,C) \\[-2pt] &\text{with }F_0=F,\ F_q=G,\text{ and, for every }0\leq j<q,\\ &F_j\rightsquigarrow F_{j+1}\quad\text{or}\quad F_{j+1}\rightsquigarrow F_j. \end{split}(8)✦

In the paper, the result reads: with r=m−kr=m-k and ε∈{1,−1}\varepsilon\in\{1,-1\}, let ℛ={(W,E)∈U(m)×Gr⁡k(ℂm):W|E=IE}\mathcal R=\{(W,E)\in U(m)\times\operatorname{Gr}_k(\mathbb C^m): W|_E=I_E\}, where Gr⁡k(ℂm)\operatorname{Gr}_k(\mathbb C^m) is the Grassmannian of kk-planes; the map G0:U(k)×ℛ→U(m)G_0:U(k)\times\mathcal R\longrightarrow U(m) is G0(X,W,E)=(Xε⊕(−Ir))W−1G_0(X,W,E)=(X^\varepsilon\oplus(-I_r))W^{-1}, with Xε=XX^\varepsilon=X for ε=1\varepsilon=1 and Xε=X*X^\varepsilon=X^* for ε=−1\varepsilon=-1, and its domain and target have real dimension m2m^2.

What Is Checked

The constructions are stated over finite complex matrix algebras and their real skew-adjoint subspaces, with smooth homotopies defined for every real parameter and endpoints prescribed on unitary elements.