OctoInformBeta

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

§3 (proof of Lemma 3.1): the regular fiber and intersection argument: The horizontal incidence differential and a fixed space bound for paired unitary maps

A regular fiber yields fixed vectors.

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

The finite-dimensional fixed space question concerns substitution of a unitary matrix XX into a smooth matrix map, followed by multiplication by a prescribed unitary matrix. A fixed space of a matrix UU is the vector space ker⁡(U−I)\ker(U-I).

Let KK, RR and LL be finite index sets, and put k=|K|k=|K| and r=|R|r=|R|. Write MI=ℂI×IM_I=\mathbb C^{I\times I} for the complex square matrices indexed by II, and U(I)\mathrm U(I) for their unitary group.✦ The skew-Hermitian space 𝔲(I)\mathfrak u(I) and the horizontal tangent space ℋK,R\mathcal H_{K,R} are the real vector spaces in (1); the rectangular matrix space is also regarded as a real vector space. 𝔲(I)={S∈MI:S*=−S},ℋK,R=𝔲(K)×𝔲(R)×ℂR×K.\mathfrak u(I)=\{S\in M_I:S^*=-S\},\qquad \mathcal H_{K,R}=\mathfrak u(K)\times\mathfrak u(R) \times\mathbb C^{R\times K}.(1)

A smooth unitary map F:MK→MK⊕RF:M_K\to M_{K\oplus R} is a map smooth over ℝ\mathbb R that sends unitary matrices to unitary matrices.✦ An elementary smooth homotopy from FF to GG is a smooth map H:ℝ×MK→MK⊕RH:\mathbb R\times M_K\to M_{K\oplus R} such that H(t,X)H(t,X) is unitary for every real tt and unitary XX, with H(0,X)=F(X)H(0,X)=F(X) and H(1,X)=G(X)H(1,X)=G(X) for such XX.✦ Smooth unitary homotopy of smooth unitary maps from MKM_K to MK⊕RM_{K\oplus R} is the equivalence relation generated by elementary smooth homotopies, allowing finite chains traversed in either direction.✦

The block diagonal matrix diag⁡(S,T)\operatorname{diag}(S,T) has diagonal blocks S,TS,T and zero off-diagonal blocks.✦ The triple matrix algebra is 𝒜K,R=MK×MK⊕R×MR\mathcal A_{K,R}=M_K\times M_{K\oplus R}\times M_R, with componentwise multiplication; its unitary group is denoted by 𝒰K,R\mathcal U_{K,R}.✦ For a family f:ℝ×MK→MK⊕Rf:\mathbb R\times M_K\to M_{K\oplus R}, its incidence map Φf\Phi_f is the matrix-valued map defined in (2). Φf(t;X,Q,T)=f(t,X)Qdiag⁡(IK,T*)Q*,(X,Q,T)∈𝒜K,R.\Phi_f(t;X,Q,T) =f(t,X)Q\operatorname{diag}(I_K,T^*)Q^*, \qquad (X,Q,T)\in\mathcal A_{K,R}.(2)✦

The model unitary map M:MK→MK⊕RM:M_K\to M_{K\oplus R} is the map in (3), and its constant family f0:ℝ×MK→MK⊕Rf_0:\mathbb R\times M_K\to M_{K\oplus R} is given there as well. For a smooth unitary map F:MK→MK⊕RF:M_K\to M_{K\oplus R}, its time-independent representative F̃:ℝ×MK→MK⊕R\widetilde F:\mathbb R\times M_K\to M_{K\oplus R} is the family in (3). For a smooth unitary map Y:MK→MLY:M_K\to M_L, the paired unitary map 𝒫Y\mathcal P_Y is defined by (4). M(X)=diag⁡(X,−IR),f0(t,X)=M(X),F̃(t,X)=F(X).M(X)=\operatorname{diag}(X,-I_R),\qquad f_0(t,X)=M(X),\qquad \widetilde F(t,X)=F(X).(3) 𝒫Y(X)=diag⁡(X,diag⁡(Y(X),Y(X)*)).\mathcal P_Y(X)=\operatorname{diag} \bigl(X,\operatorname{diag}(Y(X),Y(X)^*)\bigr).(4)✦

Writing 𝔲(𝒜K,R)\mathfrak u(\mathcal A_{K,R}) for the real vector space of triples of skew-Hermitian matrices, the off-diagonal skew-Hermitian matrix O(Z)O(Z) for Z∈ℂR×KZ\in\mathbb C^{R\times K} and the horizontal frame inclusion ι:ℋK,R→𝔲(𝒜K,R)\iota:\mathcal H_{K,R}\to\mathfrak u(\mathcal A_{K,R}) are defined by (5). O(Z)=(0−Z*Z0),ι(A,B,Z)=(A,O(Z),B).O(Z)=\begin{pmatrix}0&-Z^*\\ Z&0\end{pmatrix},\qquad \iota(A,B,Z)=(A,O(Z),B).(5)

For a real number ε\varepsilon, write Dε:ℋK,R→𝔲(K⊕R)D_\varepsilon:\mathcal H_{K,R}\to\mathfrak u(K\oplus R) for the real linear incidence differential specified in (10). The first and second diagonal blocks of S∈𝔲(K⊕R)S\in\mathfrak u(K\oplus R) are denoted by S11∈𝔲(K)S_{11}\in\mathfrak u(K) and S22∈𝔲(R)S_{22}\in\mathfrak u(R), and its lower-left block by S21∈ℂR×KS_{21}\in\mathbb C^{R\times K}. For real ε\varepsilon, the inverse differential Dε−1D_\varepsilon^{-1} is the real linear map from 𝔲(K⊕R)\mathfrak u(K\oplus R) to ℋK,R\mathcal H_{K,R} given by (6) when ε≠0\varepsilon\ne0; at ε=0\varepsilon=0, its first component is zero and its other two components are unchanged. Dε−1(S)=(ε−1S11,−S22,−12S21),ε≠0.D_\varepsilon^{-1}(S) =\bigl(\varepsilon^{-1}S_{11},-S_{22},-\tfrac12 S_{21}\bigr), \qquad \varepsilon\ne0.(6)✦

Unitary conjugation CUC_U on 𝔲(I)\mathfrak u(I), for U∈U(I)U\in\mathrm U(I), is the real linear isomorphism CU(S)=U*SUC_U(S)=U^*SU.✦ The block diagonal gauge frame is ΓU,V=diag⁡(U,V)\Gamma_{U,V}=\operatorname{diag}(U,V) for U∈U(K)U\in\mathrm U(K) and V∈U(R)V\in\mathrm U(R).✦ For U∈U(K)U\in\mathrm U(K) and V∈U(R)V\in\mathrm U(R), the model horizontal isomorphism EU,VE_{U,V} is the real linear isomorphism from ℋK,R\mathcal H_{K,R} to 𝔲(K⊕R)\mathfrak u(K\oplus R) in (7). For a real basis b=(bj)j∈Jb=(b_j)_{j\in J} of ℋK,R\mathcal H_{K,R}, the model volume basis cbc_b is the real basis of 𝔲(K⊕R)\mathfrak u(K\oplus R) given in (7). EU,V=CΓU,V*∘D1∘(CU×id),cb=(EIK,IR(bj))j∈J,E_{U,V}=C_{\Gamma_{U,V}^*}\circ D_1\circ(C_U\times\mathrm{id}), \qquad c_b=(E_{I_K,I_R}(b_j))_{j\in J},(7) where the identity acts on 𝔲(R)×ℂR×K\mathfrak u(R)\times\mathbb C^{R\times K}.

The skew-Hermitian projection π\pi from MK⊕RM_{K\oplus R} to 𝔲(K⊕R)\mathfrak u(K\oplus R) is the real linear map π(S)=12(S−S*)\pi(S)=\tfrac12(S-S^*).✦ For a real basis c=(cj)j∈Jc=(c_j)_{j\in J} of 𝔲(K⊕R)\mathfrak u(K\oplus R), the unitary tangent volume form ωc\omega_c is the real alternating form on MK⊕RM_{K\oplus R} given on families (Vj)j∈J(V_j)_{j\in J} by (8), where det⁡c\det_c is the determinant of the coordinate columns in cc. Writing p=(X,Q,T)p=(X,Q,T) for the matrix triple in (2), the horizontal incidence Jacobian Jb,cJ_{b,c} is defined by (9). ωc((Vj)j∈J)=detc((π(Vj))j∈J).\omega_c((V_j)_{j\in J})=\det_c((\pi(V_j))_{j\in J}).(8)✦ Jb,c(f;t,p)=ωc(((dΦf)(t,p)(0,pι(bj)))j∈J).J_{b,c}(f;t,p)=\omega_c\left( \bigl((d\Phi_f)_{(t,p)}(0,p\,\iota(b_j))\bigr)_{j\in J}\right).(9)✦

The Claim

Theorem 1. For real ε\varepsilon and (A,B,Z)∈ℋK,R(A,B,Z)\in\mathcal H_{K,R}, the incidence differential has the block matrix (10). When ε≠0\varepsilon\ne0, it gives the real linear isomorphism (11).✦ Dε(A,B,Z)=(εA2Z*−2Z−B).D_\varepsilon(A,B,Z)= \begin{pmatrix}\varepsilon A&2Z^*\\-2Z&-B\end{pmatrix}.(10)✦ Dε:ℋK,R→≅ℝ𝔲(K⊕R).D_\varepsilon:\mathcal H_{K,R} \xrightarrow{\ \cong_{\mathbb R}\ }\mathfrak u(K\oplus R).(11)✦

For X∈MKX\in M_K, Q∈U(K⊕R)Q\in\mathrm U(K\oplus R) and T∈U(R)T\in\mathrm U(R), the model fiber is characterized by (12). diag⁡(X,−IR)Q=Qdiag⁡(IK,T)⇔X=IK,T=−IR,Q=ΓU,V for some U∈U(K),V∈U(R).\begin{aligned} \operatorname{diag}(X,-I_R)Q&=Q\operatorname{diag}(I_K,T)\\ &\Longleftrightarrow\quad X=I_K,\quad T=-I_R,\quad Q=\Gamma_{U,V}\text{ for some }U\in\mathrm U(K),\ V\in\mathrm U(R). \end{aligned}(12)✦ For every real basis bb of ℋK,R\mathcal H_{K,R}, every t∈ℝt\in\mathbb R and every p∈𝒰K,Rp\in\mathcal U_{K,R} with Φf0(t;p)=IK⊕R\Phi_{f_0}(t;p)=I_{K\oplus R}, the horizontal Jacobian satisfies (13). 0<Jb,cb(f0;t,p).0<J_{b,c_b}(f_0;t,p).(13)✦

Suppose KK is nonempty. If a smooth unitary map F:MK→MK⊕RF:M_K\to M_{K\oplus R} is smoothly homotopic to MM, then its incidence map has the intersection (14). ∃p∈𝒰K,RΦF̃(0;p)=IK⊕R.\exists p\in\mathcal U_{K,R}\qquad \Phi_{\widetilde F}(0;p)=I_{K\oplus R}.(14)✦ For every smooth unitary map Y:MK→MLY:M_K\to M_L and every P∈U(K⊕(L⊕L))P\in\mathrm U(K\oplus(L\oplus L)), there is a unitary substitution with the fixed space bound (15). ∃X∈U(K)k≤dim⁡ℂker⁡(𝒫Y(X)P−IK⊕(L⊕L)).\exists X\in\mathrm U(K)\qquad k\le\dim_{\mathbb C}\ker\bigl(\mathcal P_Y(X)P-I_{K\oplus(L\oplus L)}\bigr).(15)✦

In the paper, the result reads: at the unique model preimage (IK,IK⊕(−IR),E0)(I_K,I_K\oplus(-I_R),E_0), where E0=ℂK⊕0E_0=\mathbb C^K\oplus0 and m=k+rm=k+r, the differential sends (A,B,Z)(A,B,Z) to (εA2Z*−2Z−B)\left(\begin{smallmatrix}\varepsilon A&2Z^*\\-2Z&-B\end{smallmatrix}\right), an isomorphism of real vector spaces; the model has degree one modulo two, so the homotopic map G(X,W,E)=F(X)W−1G(X,W,E)=F(X)W^{-1}, with WW a unitary matrix and EE a kk-plane, meets ImI_m, yielding a fixed kk-plane.

What Is Checked

The statements are over complex matrix algebras on finite index sets, with real horizontal tangent spaces as in (1) and ambient smooth unitary maps as used in (3); nonemptiness of KK is required for (14) and (15). For a nonzero real parameter, composing the incidence differential with the inverse differential in either order gives the identity on the respective real vector space.