Factorization of anti-linear and C-normal operators

(?) \implies(?) Let C be a conjugation on \mathcal{H}. Set A=T C and B=S C. Then A and B are linear operators. Since C \mathcal{H}=\mathcal{H}, we have

Then by Douglas theorem, there is a D \in \mathcal{B}(\mathcal{H}) such that

Note that R\in \mathcal{B}(\mathcal{H}).
(?)\implies (?) Using the fact that C\mathcal{H}=\mathcal{H}, we get T\mathcal{H}=S R \mathcal{H} \subseteq S \mathcal{H} and this implies R(T) \subseteq R(S).
(?) \implies (?) Observe that A A^{*}=T C C T^{\#}=T T^{\#} and B B^{*}=S S^{\#}. Therefore, T T^{\#} \leq S S^{\#} implies A A^{*} \leq B B^{*}. By Douglas theorem, A=BD and hence T=S R.
(?) \implies (?) If T=SR, then

Thus T T^{\#} \leq S S^{\# }.

Since \ker D=\ker A and \overline{\text{ran}}(D)=(\ker B)^{\perp}, we have

Thus

and

This completes the proof.

Let C be a conjugation. Then S^{\#} S=T^{\#} T implies C S^{\#} S C=C T^{\#} T C. Now set A=C S^{*} and B=C T^{*}. Then A A^{*}=B B^{*}. By Douglas Theorem,

where D is a linear partial isometry with initial Space \overline{\text{ran}}\left(B^{*}\right) and final space \overline{\text{ran}}\left(A^{*}\right).

This implies

Since \overline{\text{ran}}\left(B^{*}\right)=\ker(B)^{\perp}=\ker\left(C T^{*}\right)^{\perp}=\ker\left(T^{*}\right)^{\perp}=\overline{\text{ran}}\left(T\right) and \overline{\text{ran}}\left(A^{*}\right)=\overline{\text{ran}}(S). The operator D is linear partial isometry with initial space \overline{\text{ran}}(T) and final space \overline{\text{ran}}(S).

Since A is an anti-linear operator and C is a conjugation, the operator CA is bounded linear. Then note that |CA||CA|=(CA)^*(CA) and |A|=|CA|. Set \tilde{A}=(CA)^* and \tilde{B}=|A| and then by applying Douglas theorem, we get

where J=CD^* for some D\in \mathcal{B}(\mathcal{H}) with \ker D=\ker \tilde{A} and \overline{\text{ran}}D=(\ker \tilde{B})^\perp. Therefore,

Also, J is anti-linear and isometry on \overline{\text{ran}}(|A|)=(\ker J)^\perp.

Let T\in \mathcal{B}(\mathcal{H}) be C-normal operator that is, CT^*TC=TT^* equivalently, C|T|C=|T^*|. Then by setting A=CT^*C and B=|T^*|, we have AA^*=BB^*. Now apply Douglas theorem, we get A=BD^* for some D:\overline{\text{ran}}(B^*)\rightarrow\overline{\text{ran}}(A^*) such that DB^*x=A^*x, and Dx=0 for all x\in \overline{\text{ran}}(B^*)^\perp=\ker B. Thus we have

By defininig J=DC, we get T=CJ|T|. Note that

and

Since \overline{\text{ran}}(B^*)=\overline{\text{ran}}(|T^*|)=C\overline{\text{ran}}(|T|), and \overline{\text{ran}}(A^*)=\overline{\text{ran}}(|T|), it is evident that J=DC is anti-linear unitary operator supported on \overline{\text{ran}}(|T|).

Also,

This implies |T^*|=CJ|T|J^\#C and consequently, J|T|=|T|J.

The converse follows trivially.

The proof follows from Proposition 2.8.

Since T is normal, \ker T is a reducing subspace for T (cf. [Conway:Book:Functional, analysis, Proposition 5.6]). Hence by Lemma 4.1, the conclusion follows.

Since T is normal, we have \ker T=\ker T^*. Write \mathcal{H}=\ker T\oplus(\ker T)^\perp. Then T can be written as T= \begin{bmatrix} 0&0\\ 0&A \end{bmatrix}, where A:(\ker T)^\perp\rightarrow(\ker T)^\perp.

Now T is C-normal which implies C\ker T=\ker T^*=\ker T. Therefore, C can be written as C=\begin{bmatrix} C_1&0\\ 0&C_2 \end{bmatrix}, where C_1:\ker T\rightarrow\ker T and C_2:(\ker T)^\perp\rightarrow(\ker T)^\perp. Note that C_2 is also a conjugation on (\ker T)^\perp and CT^*TC=TT^* implies that C_2A^*AC_2=AA^* which implies A:=T|_{(\ker T)^\perp} is C_2-normal.

Since T is C-normal, the operator S_2 satisfies CS_2C=S_2. Hence, by using [Wang:Xie:Yan:Zhu, Theorem 5], the conclusion follows.

Since CTC=-T, we have \sigma(T)=-\sigma(T)^*. Let f\in C(\sigma(T)), the space of all continuous functions defined on \sigma(T), define f^*(z)=\overline{f(-\bar{z})} for all z\in \sigma(T). Then the function f^*\in C(\sigma(T)). For g(z)=\alpha z^n\bar{z}^m, it follows that

Since all bivariate polynomials in z and \bar{z} are uniformly dense in C(\sigma(T)), it follows that

Let \chi_{\Delta} be the characteristic function of \Delta belongs to L^\infty(\mu), where \mu is a scalar-valued spectral measure for T. Thus there exits a sequence \{f_n\} in C(\sigma(T)) such that f_n\rightarrow\chi_{\Delta} and f^*_n\rightarrow\chi_{-\Delta^*} in the weak^*-topology of L^\infty(\mu). Thus by functional calculus for normal operators (cf. [Conway:Book:Functional, analysis, Theorem 8.10]), we get f_n(T)\rightarrow\chi_{\Delta}(T) and f^*_n(T)\rightarrow\chi_{-\Delta^*}(T) in the weak operator topology. Therefore, we obtain C\chi_{\Delta}(T)C=\chi_{-\Delta^*}(T), that is, CE_T(\Delta)C=E_T(-\Delta^*), and hence the desired conclusion follows.

Denote \Delta_1=\Sigma_1\cap \sigma(T) and \Delta_2=\Sigma_2\cap \sigma(T). Then \sigma(T)=\Delta_1\cup (-\Delta_1^*)\cup \Delta_2. Denote \mathcal{H}_1=\text{ran} E_T(\Delta_1), \mathcal{H}_2=\text{ran}E_T(-\Delta_1^*), and \mathcal{H}_3=\text{ran}E_{T}(\Delta_2). Then \mathcal{H}=\mathcal{H}_1\oplus\mathcal{H}_2\oplus\mathcal{H}_3. Since T is normal, the subspaces \mathcal{H}_1,\mathcal{H}_2,\mathcal{H}_3 reduces T. Then T=T_1\oplus T_2\oplus T_3, where T_i =P_{\mathcal{H}_i}T|_{\mathcal{H}_i}. By Lemma 4.6, we have C\mathcal{H}_1=\mathcal{H}_2 and C\mathcal{H}_2=\mathcal{H}_1. Therefore, we have

where D:\mathcal{H}_1\rightarrow\mathcal{H}_2 is an anti-linear, invertible, isometry, and C_3:\mathcal{H}_3\rightarrow\mathcal{H}_3 is a conjugation. Since T satisfies CTC=-T, we have T_2=-DT_1D^{-1}. Again, T_1 is normal, so there exists a conjugation C_1 on \mathcal{H}_1 such that C_1T_1C_1=T_1^*, and hence we have T_2=-DC_1T_1^*C_1D^{-1}. Observe that DC_1 is an unitary operator. Thus T_2\cong -T_1^* (unitarily equivalent).

Proof of (?) Since T is C-normal as well as normal, we have

which shows that C commutes with T^*T. Then C commutes with f(T^*T) for every f\in C(\sigma(T^*T)). Let G be a nonempty open subset of \sigma(T^*T). Let \{f_n\}\subseteq C(\sigma(T^*T)) be such that 0\leq f_n\leq \chi_G and f_n\nearrow \chi_G. Then

and it follows that

This shows that C commutes with spectral measure of T^*T.

Proof of (?): Since T is C-normal, then we have

which shows that CT commutes with T^*T. Therefore, following the same steps as in (?), we get the desired conclusion.

Since the linear operator T^*T is positive, by spectral theorem for normal operator, there is a unitary operator U:\mathcal{H}\rightarrow L^2(\mu) such that T^*T=U^*M_xU, where M_x is the multiplication operator induced by the real variable x. Therefore, E_{T^*T}=U^*E_{M_x}U, where E_{T^*T} and E_{M_x}are the projection-valued spectral measures corresponding to T^*T and M_x, respectively. Since T is C-normal and normal, by (?) of Theorem 4.9, we have

Observe that S_\mu M_x=M_xS_\mu, as well as S_\mu E_{M_x}=E_{M_x}S_\mu. Therefore, by Proof Proof, we have

and this implies the linear operator UCU^*S_\mu belongs to the commutant of M_x.

Hence there exists an essentially bounded measurable function \varphi such that UCU^*S_\mu=M_\varphi, where M_\varphi is the multiplication operator on L^2(\mu) induced by \varphi with |\varphi|=1 \mu-a.e. It follows that

Hence by using Remark 4.11, we obtain the desired conclusion.

Let C be a conjugation on \mathcal{H}. In (?) of Theorem 4.9, we have seen that if T is C-normal, then the anti-linear operator CT commutes with the projection-valued spectral measure E_{T^*T} associated to the positive operator T^*T. By continuing, in the similar fashion as earlier, we have

Hence the linear operator UCTU^*S_\mu is a commutant of M_x. Therefore, there exists \varphi\in L^\infty(\mu) such that UCTU^*S_\mu=M_\varphi, which implies that T=CU^*M_\varphi S_\mu U.

Let C be a conjugation, and T=A+iB, where A is C-symmetric and B is C-skew-symmetric (cf. [RSV_Cart, Theorem 2.3]). Therefore, for every x\in \mathcal{H}

and

Since T is C-normal, we have \left\|Tx\right\|=\left\|T^*Cx\right\|, and this implies

Thus for every x\in \mathcal{H}, we have T is C-normal if and only if \left\|Tx\right\|=\left\|T^*Cx\right\| if and only if \Re (i\langle Ax,Bx\rangle)=0 if and only if \left\|Tx\right\|^2=\left\|Ax\right\|^2+\left\|Bx\right\|^2.

Now observe that \Re (i\langle Ax,Bx\rangle)=0 if and only if \langle Ax,Bx\rangle\in \mathbb{R} for all x\in \mathcal{H} if and only if (B^*A)^*=B^*A, that is, A^*B=B^*A. Since A is C-symmetric and B is C-skew-symmetric, we have A^*B=B^*A\iff CA^*B=CB^*A \iff ACB=-BCA \iff -AB^*C=-BA^*C\iff AB^*=BA^*. This completes the proof.

To prove this, we use the Cartesian decomposition of C-normal operators (See Theorem 5.1). We write T=A+iB, where

and

Also note that

and

The condition AB^*=BA^* implies

and this implies |\lambda_{n-j}|=|\lambda_j| for all 1\leq j\leq n-1. Hence the conclusion follows.

Since T is a compact C-normal operator, it follows from Theorem 5.1 that T=A+iB, where A is C-symmetric and B is C-skew-symmetric such that A^*B=B^*A and |T|=\sqrt{|A|^2+|B|^2} with |A||B|=|B||A|. Thus we have

Hence, by using Weyl’s monotonicity principle the conclusion follows.

It is well known that if A,B are compact operators, then

for j=1,2,\dots (cf. [Bhatia:Kittaneh]). Therefore, if T\in \mathcal{B}(\mathcal{H}) is compact C-normal with the Cartesian decomposition T=A+iB, by using Theorem 5.1, we have T^*T=A^*A+B^*B and TT^*=AA^*+BB^*, and hence by Proof Proof, we have 2s_j(BA^*)=2s_j(AB^*)\leq s_j((T^*T)\oplus(TT^*) ) for j=1,2,\dots.

The desired result follows easily from a direct calculation.