Abstract
Let H and K be infinite dimensional separable complex Hilbert spaces and B(K, H) the algebra of all bounded linear operators from K into H. Let \(A\in B(H)\) and \(B\in B(K)\). We denote by \(M_C\) the operator acting on \(H\oplus K\) of the form \(M_C=\left( \begin{array}{cc}A&{}C\\ 0&{}B\\ \end{array}\right) \). In this paper, we give necessary and sufficient conditions for \(M_C\) to be an upper semi-Fredholm operator with \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\) for some left invertible operator \(C\in B(K,H)\). Meanwhile, we discover the relationship between \(n(M_C)\) and n(A) during the exploration. And we also describe all left invertible operators \(C\in B(K,H)\) such that \(M_C\) is an upper semi-Fredholm operator with \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\).
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Throughout this paper, \(\mathbb {C}\) and \(\mathbb {N}\) denote the set of complex numbers and the set of nonnegative integers, respectively. Let H and K be complex separable infinite dimensional Hilbert spaces and B(K, H) the algebra of all bounded linear operators from K into H. If \(K=H\), then we denote B(H, H) by B(H). Let \(T\in B(H)\). We denote by n(T) the nullity of T, which is the dimension of the kernel N(T), and by d(T) the codimension of the range R(T). If R(T) is closed and \(n(T)<\infty \), T is called an upper semi-Fredholm operator (or left semi-Fredholm operator). T is said to be a lower semi-Fredholm operator (or right semi-Fredholm operator) if \(d(T)<\infty \)[1, Corollary 1.15]. An operator T is said to be a Fredholm operator if it is both lower and upper semi-Fredholm. Especially, if T is an upper semi-Fredholm operator and \(n(T)=0\), then T is called a left invertible operator. Put \(B_{l}^{-1}(K, H)=\{T\in B(K, H): T\ \hbox {is\ a\ left\ invertible\ operator}\}\). If T is a lower or upper semi-Fredholm operator, the index of T is defined by \(\hbox {ind}(T)=n(T)-d(T)\). An operator T is said to be an upper semi-Weyl operator if it is an upper semi-Fredholm operator with \(\hbox {ind}(T)\le 0\). If T is a Fredholm operator and \(\hbox {ind}(T)=0\), then T is called a Weyl operator. Let \(T^*\) denote the conjugate of \(T\in B(H)\).
Let \(A\in B(H)\), \(B\in B(K)\) and \(C\in B(K,H)\). We define a \(2\times 2\) upper triangular operator matrix \(M_C\) acting on \(H\oplus K\) of the form
The completion problems of operator matrices were considered for many types of matrices, for example, \(M_C\), \(M_{X}\) and \(M_{(X,Y)}\). In the past 20 years, the study of completion problems for upper triangular operator matrices has been an attractive subject for many scholars and many valuable results have been obtained [2,3,4,5,6]. In [7], the authors gave the necessary and sufficient conditions for \(M_C\) to be an upper semi-Weyl operator for some operator \(C\in B(K,H)\). Later, Wu and Huang et al. studied the self-adjoint perturbations of spectra for operator matrices in [8, 9]. In recent years, Dong and Cao gave the necessary and sufficient conditions which make \(M_C\) be a CFI operator for some operator \(C\in B(K,H)\) in [10].
The idea of our article comes from the following papers:
In [11], the authors have shown that there exists some operator \(C\in B(K,H)\) such that \(M_C\) is upper semi-Fredholm and \(\hbox {ind}(M_C)\le 0\) if and only if there exists some left invertible operator \(C\in B(K,H)\) such that \(M_C\) is upper semi-Fredholm and \(\hbox {ind}(M_C)\le 0\).
In [2], the authors briefly discuss the deficiency and the nullity of upper triangular operator matrices in the process of proving the left and right invertibility of \(M_X\).
In this paper, we further give necessary and sufficient conditions for \(M_C\) to be an upper semi-Fredholm operator, \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\) for some left invertible operator \(C\in B(K,H)\) by different methods. During the discussion, we also found a new relationship between \(n(M_C)\) and n(A), which further reflects the fixed nullity problem of operator matrices. The main results in this paper are the following theorems:
Theorem 1
Suppose that \(A\in B(H)\) and \(B\in B(K)\) are given. Then \(M_C\) is an upper semi-Fredholm operator, \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\) for some left invertible operator \(C\in B(K,H)\) if and only if A is an upper semi-Fredholm operator, \(n(A)+n(B)>0\) and one of the following statements holds:
-
(1)
B is an upper semi-Fredholm operator and \(n(A)+n(B)<d(A)+d(B)\);
-
(2)
\(d(A)=\infty \).
Theorem 2
Let \(A\in B(H)\) and \(B\in B(K)\). Then, \(M_C\) is an upper semi-Fredholm operator, \(n(M_C)=\hbox {max}\{n(A), 1\}\) and \(\hbox {ind}(M_C)<0\) for some left invertible operator \(C\in B(K,H)\) if and only if A is an upper semi-Fredholm operator, \(n(A)+n(B)>0\) and one of the following statements holds:
-
(1)
B is upper semi-Fredholm, \(n(A)+n(B)<d(A)+d(B)\) and \(n(A)>0\), \(n(B)\le d(A)\);
-
(2)
B is upper semi-Fredholm, \(n(A)+n(B)<d(A)+d(B)\) and \(n(A)=0\), \(n(B)\le d(A)+1\);
-
(3)
\(d(A)=\infty \).
Furthermore, we will discuss all \(C\in B_{l}^{-1}(K, H)\) such that \(M_C\) is upper semi-Fredholm, \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\) by Theorem 2. We get the following result:
Theorem 3
Let \(A\in B(H)\) and \(B\in B(K)\). Then, \(M_C\) is an upper semi-Fredholm operator, \(n(M_C)=\hbox {max}\{n(A), 1\}\) and \(\hbox {ind}(M_C)<0\) for all left invertible operators \(C\in B(K,H)\) if and only if one of the following statements holds:
-
(1)
A is upper semi-Fredholm with \(n(A)>0\), B is left invertible and \(n(A)+n(B)<d(A)+d(B)\);
-
(2)
A is invertible, B is upper semi-Fredholm with \(n(B)=1\) and \(n(A)+n(B)<d(A)+d(B)\).
2 Auxiliary Results
In order to prove Theorem 1, we give some useful lemmas.
Lemma 1
Let \(A\in B(H)\) be a left invertible operator with \(d(A)=\infty \) and \(B\in B(K)\) with \(n(B)>0\), then there exists some left invertible operator \(C\in B(K,H)\) such that \(M_C\) is an upper semi-Fredholm operator, \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\).
Proof
We take \(e_1\in N(B)\) with \(\Vert e_1\Vert =1\) and extend it into an orthonormal basis of K, written as \(\{e_{n}\}_{n=1}^{\infty }\). Put \(M=\hbox {span}\{e_1\}\). Then \(\hbox {dim}M^{\perp }=\infty =d(A)=\hbox {dim} R(A)^{\perp }\). So, there is an isometric invertible linear operator \(T_1:M^{\perp }\rightarrow R(A)^{\perp }\). Take \(u_1=Aw_1\in R(A)\) with \(\Vert u_1\Vert =1\), and let \(N=\hbox {span}\{u_1\}\). Let \(T_2\) be an isometric invertible linear operator as \(T_2: M\rightarrow N\) satisfying \(T_2e_1=u_1=Aw_1\). Define \(C: K\rightarrow H\) by
It is clear that \(C\in B(K,H)\) is left invertible. We claim that \(M_C\) is an upper semi-Fredholm operator with \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\).
-
(i)
\(0<n(M_C)=1<\infty \).
Let \(\left( \begin{array}{c}x \\ y \\ \end{array}\right) \in N(M_C)\). Then \(Ax+Cy=0\) and \(By=0\). Let \(y=\alpha _1+k_1e_1\), where \(k_1\in \mathbb {C}\) and \(\alpha _1\in M^{\perp }\). It follows from \(Ax+Cy=Ax+T_1\alpha _1+T_2(k_1e_1)=A(x+k_1w_1)+T_1\alpha _1\) that \(A(x+k_1w_1)=-T_1\alpha _1\in R(A)\cap R(A)^{\perp }=\{0\}\). From \(T_1\) is invertible and A is left invertible, we can get that \(\alpha _1=0\), \(x=-k_1w_1\) and \(y=k_1e_1\). Hence, \(N(M_C)\subseteq \hbox {span}\left\{ \left( \begin{array}{ccc} w_1 \\ -e_1 \\ \end{array}\right) \right\} \). Since the inclusion “\(\supseteq \)” is obvious, we get that \(N(M_C)=\hbox {span}\left\{ \left( \begin{array}{ccc} w_1 \\ -e_1 \\ \end{array}\right) \right\} \). Therefore, \(0<n(M_C)=1<\infty \).
-
(ii)
\(R(M_C)\) is closed.
Suppose that \(M_C\left( \begin{array}{c}x_{n} \\ y_{n} \\ \end{array}\right) \rightarrow \left( \begin{array}{c}u_{0} \\ v_{0} \\ \end{array}\right) \) (\(n\rightarrow \infty \)). Then, \(Ax_{n}+Cy_{n}\rightarrow u_{0}\) and \(By_{n}\rightarrow v_{0}\) (\(n\rightarrow \infty \)). Let \(y_n=\alpha _n+k_ne_1\), where \(k_n\in \mathbb {C}\) and \(\alpha _n\in M^{\perp }\). So \(Ax_n+Cy_n=Ax_n+T_1\alpha _n+T_2(k_ne_1)=A(x_n+k_nw_1)+T_1\alpha _n\rightarrow u_{0}\) (\(n\rightarrow \infty \)). Since \(A(x_n+k_nw_1)\in R(A)\) and \(T_1\alpha _n\in R(A)^{\perp }\), we get that both \(\{A(x_n+k_nw_1)\}_{n=1}^{\infty }\) and \(\{T_1\alpha _n\}_{n=1}^{\infty }\) are Cauchy sequences. From the fact that \(T_1\) is invertible, we can get that \(\{\alpha _n\}_{n=1}^{\infty }\) is a Cauchy sequence. From R(A) and \(M^{\perp }\) are closed, there exist some \(x_{0}\in H\) and \(\alpha _0\in M^{\perp }\) such that \(A(x_n+k_nw_1)\rightarrow Ax_{0}\), \(\alpha _n\rightarrow \alpha _0\) (\(n\rightarrow \infty \)). Now, \(Ax_n+Cy_n=A(x_n+k_nw_1)+C\alpha _n\rightarrow Ax_{0}+C\alpha _0\) and \(By_n=B\alpha _n\rightarrow B\alpha _0\) as \(n\rightarrow \infty \). This implies that \(M_C\left( \begin{array}{c}x_{n} \\ y_{n} \\ \end{array}\right) \rightarrow M_C\left( \begin{array}{c}x_0 \\ \alpha _0 \\ \end{array}\right) \). Therefore \(\left( \begin{array}{c}u_{0} \\ v_{0} \\ \end{array}\right) =M_{C}\left( \begin{array}{c}x_{0} \\ \alpha _{0} \\ \end{array}\right) \in R(M_{C})\), which means that \(R(M_C)\) is closed.
-
(iii)
\(\hbox {ind}(M_C)<0\).
Claim 1: \(\left( \begin{array}{c}0 \\ e_{i} \\ \end{array}\right) \notin R(M_{C})\) for any \(i\ge 2\).
In fact, if there exists some \(k\ge 2\) such that \(\left( \begin{array}{c}0 \\ e_{k} \\ \end{array} \right) \in R(M_{C})\), we may suppose that \(\left( \begin{array}{c}0 \\ e_{k} \\ \end{array}\right) =M_{C}\left( \begin{array}{c}x_0 \\ y_0 \\ \end{array}\right) \), then \(Ax_0+Cy_0=0\) and \(By_0=e_{k}\). Let \(y_0=\alpha _0+k_1e_1\), where \(k_1\in \mathbb {C}\) and \(\alpha _0\in M^{\perp }\). From \(Ax_0+Cy_0=Ax_0+T_1\alpha _0+T_2(k_1e_1)=A(x_0+k_1w_1)+T_1\alpha _0=0\) we can get that \(A(x_0+k_1w_1)=-T_1\alpha _0\in R(A)\cap R(A)^{\perp }=\{0\}\). Since \(T_1\) is invertible and A is left invertible, it follows that \(\alpha _0=0\) and \(x_0=-k_1w_1\). Then, \(y_0=k_1e_1\) and \(By_0=B(k_1e_1)=0\), a contradiction.
Claim 2: The nonzero linear combinations of \(\left\{ \left( \begin{array}{ccc} 0 \\ e_i \\ \end{array}\right) \right\} _{i=2}^\infty \) are not in \(R(M_{C})\).
In fact, if there exists \(\left( \begin{array}{c}x_1 \\ y_1 \\ \end{array} \right) \in H\oplus K\) such that \(\sum \limits _{i=2}^na_i\left( \begin{array}{c} 0 \\ e_{i} \\ \end{array}\right) =M_{C}\left( \begin{array}{c}x_1 \\ y_1 \\ \end{array} \right) \), where \(a_i\ne 0\) for some \(2\le i\le n\). Then \(Ax_1+Cy_1=0\) and \(By_1=a_2e_2+\cdots +a_ne_n\). Let \(y_1=\alpha _1+k_2e_1\), where \(k_2\in \mathbb {C}\) and \(\alpha _1\in M^{\perp }\). Similar to the proof of Claim 1, we get that \(y_1=k_2e_1\). Then, \(By_1=B(k_2e_1)=0\), a contradiction.
Claim 3: \(d(M_C)=\infty \).
We assume that \(\left( \begin{array}{c}0 \\ e_{i} \\ \end{array}\right) =g_i+h_i\) for any \(i\ge 2\), where \(g_i\in R(M_C)\) and \(h_i\in R(M_C)^{\perp }\). For any \(n\ge 2\), if \(\sum \limits _{i=2}^na_ih_i=0\), then \(a_2\left( \begin{array}{c}0 \\ e_{2} \\ \end{array}\right) \)+\(\cdots \)+\(a_n\left( \begin{array}{c}0 \\ e_{n} \\ \end{array}\right) =a_2g_2+\cdots +a_ng_n\in R(M_C)\). From Claim 2 we know that \(a_2\left( \begin{array}{c}0 \\ e_{2} \\ \end{array}\right) \)+\(\cdots \)+\(a_n\left( \begin{array}{c}0 \\ e_{n} \\ \end{array}\right) =0\). Since \(\{e_i\}_{i=2}^{\infty }\) is linearly independent, we get that \(a_i=0\), \(2\le i\le n\) and so \(\{h_i\}_{i\ge 2}\) is linearly independent. It follows that \(\hbox {dim}R(M_C)^{\perp }=d(M_C)=\infty \), which implies \(\hbox {ind}(M_C)<0\).
From Lemma 1, if \(A\in B(H)\) is a left invertible operator with \(d(A)=\infty \) and \(B\in B(K)\) with \(n(B)>0\), then there exists some left invertible operator \(C\in B(K,H)\) such that \(M_C\) is an upper semi-Fredholm operator, \(n(M_C)=1\) and \(\hbox {ind}(M_C)<0\). If \(A\in B(H)\) is upper semi-Fredholm with \(n(A)>0\), we can get that:
Lemma 2
Let \(A\in B(H)\) and \(B\in B(K)\). If A is an upper semi-Fredholm operator with \(d(A)=\infty \) and \(n(A)>0\), then there exists \(C\in B_{l}^{-1}(K, H)\) such that \(M_C\) is an upper semi-Fredholm operator, \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\).
Proof
We will complete the proof by the following two cases.
Case 1. R(B) is not closed.
There exists an isometric invertible linear operator \(T:\ K\rightarrow R(A)^{\perp }\) since \(\dim K=\dim R(A)^{\perp }=\infty \).
Define an operator \(C:\ K\rightarrow H\) by \(Cy=Ty\) for any \(y\in K\). It is clear that C is left invertible. Next we will show \(M_C\) has the desired properties.
-
(i)
\(0<n(M_C)=n(A)<\infty \).
Let \(\left( \begin{array}{c}x \\ y \\ \end{array}\right) \in N(M_C)\). Then \(Ax+Cy=0\) and \(By=0\). It follows that \(Ax=-Cy=-Ty\in R(A)\cap R(A)^{\perp }=\{0\}\). From T is invertible, we can get that \(y=0\) and \(x\in N(A)\). Hence, \(N(M_C)\subseteq N(A)\oplus \{0\}\). The inclusion “\(\supseteq \)" is obvious, then \(N(M_C)=N(A)\oplus \{0\}\). Therefore, \(0<n(M_C)=n(A)<\infty \).
-
(ii)
\(R(M_C)\) is closed.
Suppose that \(M_C\left( \begin{array}{c} x_{n} \\ y_{n} \\ \end{array} \right) \rightarrow \left( \begin{array}{c} u_{0} \\ v_{0} \\ \end{array} \right) \) (\(n\rightarrow \infty \)). Then \(Ax_{n}+Cy_{n}=Ax_n+Ty_n\rightarrow u_{0}\) and \(By_{n}\rightarrow v_{0}\) (\(n\rightarrow \infty \)). Since \(Cy_n=Ty_n\in R(A)^{\perp }\), we get that both \(\{Ax_n\}_{n=1}^{\infty }\) and \(\{Ty_n\}_{n=1}^{\infty }\) are Cauchy sequences. It follows that \(\{y_n\}_{n=1}^{\infty }\) is a Cauchy sequence. We suppose that \(y_n\rightarrow y_0\) (\(n\rightarrow \infty \)). Since R(A) is closed, there exists some \(x_{0}\in H\) such that \(Ax_n\rightarrow Ax_{0}\)(\(n\rightarrow \infty \)). Now, \(Ax_n+Cy_n\rightarrow Ax_{0}+Cy_0\) and \(By_n\rightarrow By_0\) (\(n\rightarrow \infty \)). Therefore, \(\left( \begin{array}{c}u_{0} \\ v_{0} \\ \end{array}\right) =M_{C}\left( \begin{array}{c}x_{0} \\ y_{0} \\ \end{array}\right) \in R(M_{C})\). It means that \(R(M_C)\) is closed.
-
(iii)
From R(B) is not closed we get that \(d(M_C)\ge d(B)=\infty \) and so \(\hbox {ind}(M_C)<0\).
Case 2. R(B) is closed.
Now, we divide it according to the nullity of B.
-
(1)
Suppose \(n(B)<\infty \).
In this case, B is an upper semi-Fredholm operator. It is obvious that \(M_C\) is an upper semi-Fredholm operator and \(\hbox {ind}(M_C)=\hbox {ind}(A)+\hbox {ind}(B)<0\) for all left invertible operators \(C\in B(K,H)\). Since \(\hbox {dim}R(A)^{\perp }=\infty \), there is an isometric invertible linear operator \(T:K\rightarrow R(A)^{\perp }\). Define an operator \(C:K\rightarrow H\) by \(Cy=Ty\) for any \(y\in K\). It is clear that C is left invertible. Similar to the proof of Case 1, \(n(M_C)=n(A)\) is trivial.
-
(2)
Suppose \(n(B)=\infty \) and \(n(A)<d(B)\).
Since \(\hbox {dim}N(B)=\hbox {dim}R(A)^{\perp }=\infty \), there exists an isometric invertible linear operator \(T_2:N(B)\rightarrow R(A)^{\perp }\). It is clear that \(\hbox {dim}R(A)=\infty \). Therefore, there is a closed subspace \(M\subseteq R(A)\) such that \(\hbox {dim}N(B)^{\perp }=\hbox {dim}M\). So, there exists an isometric invertible linear operator \(T_1:N(B)^{\perp }\rightarrow M\). Define \(C:K\rightarrow H\) by
It is clear that \(C\in B(K,H)\) is left invertible. In the following, we prove that \(M_C\) satisfies the above conditions.
-
(i)
\(0<n(M_C)=n(A)<\infty \).
Let \(\left( \begin{array}{c}x \\ y \\ \end{array}\right) \in N(M_C)\). Then \(Ax+Cy=0\) and \(By=0\). Then, \(y\in N(B)\) and \(Ax=-Cy=-T_2y\in R(A)\cap R(A)^{\perp }=\{0\}\). From \(T_2\) is invertible, we can get that \(y=0\) and \(x\in N(A)\). Hence, \(N(M_C)\subseteq N(A)\oplus \{0\}\). The inclusion “\(\supseteq \)" is obvious, then \(N(M_C)=N(A)\oplus \{0\}\). Therefore, \(0<n(M_C)=n(A)<\infty \).
-
(ii)
\(R(M_C)\) is closed.
Suppose that \(M_C\left( \begin{array}{c}x_{n} \\ y_{n} \\ \end{array}\right) \rightarrow \left( \begin{array}{c}u_{0} \\ v_{0} \\ \end{array}\right) \) (\(n\rightarrow \infty \)). Then \(Ax_{n}+Cy_{n}\rightarrow u_{0}\) and \(By_{n}\rightarrow v_{0}\) (\(n\rightarrow \infty \)). Let \(y_n=\alpha _n+\beta _n\), where \(\alpha _n\in N(B)^{\perp }\) and \(\beta _n\in N(B)\). So \(Ax_n+Cy_n=(Ax_n+T_1\alpha _n)+T_2\beta _n\rightarrow u_{0}\) (\(n\rightarrow \infty \)). Since \(Ax_n+T_1\alpha _n\in R(A)\) and \(T_2\beta _n\in R(A)^{\perp }\), we get that both \(\{Ax_n+T_1\alpha _n\}_{n=1}^{\infty }\) and \(\{T_2\beta _n\}_{n=1}^{\infty }\) are Cauchy sequences. It follows that \(\{\beta _n\}_{n=1}^{\infty }\) is a Cauchy sequence. Note the fact that \(B|_{N(B)^{\perp }}\) is left invertible and \(By_n=B\alpha _n\), then we can obtain that \(\{\alpha _n\}_{n=1}^{\infty }\) is a Cauchy sequence. Since R(A) is closed, there exist some \(x_{0}\in H\) and \(y_0\in K\) such that \(Ax_n\rightarrow Ax_{0}\), \(y_n\rightarrow y_0\) (\(n\rightarrow \infty \)). Now, \(Ax_n+Cy_n\rightarrow Ax_{0}+Cy_0\) and \(By_n\rightarrow By_0\) (\(n\rightarrow \infty \)). Therefore, \(\left( \begin{array}{c}u_{0} \\ v_{0} \\ \end{array}\right) =M_{C}\left( \begin{array}{c}x_{0} \\ y_{0} \\ \end{array}\right) \in R(M_{C})\).
-
(iii)
From \(d(M_C)\ge d(B)>n(A)=n(M_C)\) we get that \(\hbox {ind}(M_C)<0\).
-
(3)
Suppose \(n(B)=\infty \) and \(d(B)\le n(A)\).
Now, B is a lower semi-Fredholm operator and then B is not compact. So \(\hbox {dim}N(B)^{\perp }=\hbox {dim}R(B^*)=\infty \). Since \(d(B)\le n(A)<\infty \), there exists a positive integer m such that \(n(A)<d(B)+m\). Put \(N(B)^{\perp }=M\oplus M^{\perp }\), where \(\hbox {dim}M=m\). There are isometric invertible linear operators \(T_1:M^{\perp }\rightarrow R(A)\) and \(T_2:N(B)\oplus M\rightarrow R(A)^{\perp }\). Define \(C:K\rightarrow H\) by
It is clear that \(C\in B(K,H)\) is invertible and so is left invertible.
-
(i)
\(0<n(M_C)=n(A)<\infty \).
It is trivial to get this result.
-
(ii)
\(R(M_C)\) is closed.
Suppose that \(M_C\left( \begin{array}{c} x_{n} \\ y_{n} \\ \end{array} \right) \rightarrow \left( \begin{array}{c} u_{0} \\ v_{0} \\ \end{array} \right) \) (\(n\rightarrow \infty \)). Then, \(Ax_{n}+Cy_{n}\rightarrow u_{0}\) and \(By_{n}\rightarrow v_{0}\) (\(n\rightarrow \infty \)). Let \(y_n=\alpha _n+\beta _n\), where \(\alpha _n\in M^{\perp }\) and \(\beta _n\in N(B)\oplus M\). So \(Ax_n+Cy_n=(Ax_n+T_1\alpha _n)+T_2\beta _n\rightarrow u_{0}\) (\(n\rightarrow \infty \)). Since \(Ax_n+T_1\alpha _n\in R(A)\) and \(T_2\beta _n\in R(A)^{\perp }\), we get that both \(\{Ax_n+T_1\alpha _n\}_{n=1}^{\infty }\) and \(\{T_2\beta _n\}_{n=1}^{\infty }\) are Cauchy sequences. It follows that \(\{\beta _n\}_{n=1}^{\infty }\) is a Cauchy sequence. From \(By_n=B\alpha _n+B\beta _n\) we can get that \(\{B\alpha _n\}_{n=1}^{\infty }\) is a Cauchy sequence. Note the fact that \(B|_{N(B)^{\perp }}\) is left invertible and \(\{\alpha _n\}_{n=1}^{\infty }\subseteq N(B)^{\perp }\). We can obtain that \(\{\alpha _n\}_{n=1}^{\infty }\) is a Cauchy sequence, then \(\{y_n\}_{n=1}^{\infty }\) is a Cauchy sequence. Since R(A) is closed, there exist some \(x_{0}\in H\) and \(y_0\in K\) such that \(Ax_n\rightarrow Ax_{0}\), \(y_n\rightarrow y_0\) (\(n\rightarrow \infty \)). Now, \(Ax_n+Cy_n\rightarrow Ax_{0}+Cy_0\) and \(By_n\rightarrow By_0\) (\(n\rightarrow \infty \)). Therefore, \(\left( \begin{array}{c} u_{0} \\ v_{0} \\ \end{array} \right) =M_{C}\left( \begin{array}{c} x_{0} \\ y_{0} \\ \end{array} \right) \in R(M_{C})\).
-
(iii)
\(\hbox {ind}(M_C)<0\).
Let \(\{e_i\}_{i=1}^{m}\) be an orthonormal basis of M. From \(\{e_i\}_{i=1}^{m}\subseteq N(B)^{\perp }=R(B^*)\) we can suppose that \(e_i=B^*u_i\) for some \(u_i\), \(1\le i\le m\). Since \(Ce_i\in R(A)^{\perp }=N(A^*)\), we have \(A^*(Ce_i)=0\). Moreover, \(C^*(Ce_i)=C^*(T_2e_i)=T_2^*T_2e_i=e_i\), so \(\left\{ \left( \begin{array}{ccc} Ce_i \\ -u_i \\ \end{array}\right) \right\} _{i=1}^{m}\subseteq N(M_C^*)\). From the fact that \(T_2\) is invertible and \(Ce_i=T_2e_i\), we get that \(\left\{ \left( \begin{array}{ccc} Ce_i \\ -u_i \\ \end{array}\right) \right\} _{i=1}^{m}\) is linearly independent. Since \(n(B^*)=d(B)<\infty \), we assume that \(h_1, h_2, \cdots , h_{d(B)}\in N(B^*)\) are linearly independent. Then, \(\left\{ \left( \begin{array}{ccc} 0 \\ h_i \\ \end{array}\right) \right\} _{i=1}^{d(B)}\subseteq N(M_C^*)\).
We claim that: \(\left( \begin{array}{c} 0 \\ h_1 \\ \end{array} \right) , \cdots , \left( \begin{array}{c} 0 \\ h_{d(B)} \\ \end{array} \right) , \left( \begin{array}{c} Ce_1 \\ -u_1 \\ \end{array} \right) , \cdots , \left( \begin{array}{c} Ce_m \\ -u_m \\ \end{array} \right) \) are linearly independent.
In fact, we assume that
then
Since C is invertible and \(\{e_i\}_{i=1}^{m}\) is linearly independent, we obtain that \(\lambda _1=\cdots =\lambda _m=0\). Hence, \(a_1h_1+\cdots +a_{d(B)}h_{d(B)}=0\) and then \(a_1=a_2=\cdots =a_{d(B)}=0\). So this claim is proved.
Therefore, \(n(M_C^*)\ge d(B)+m\) and \(\hbox {ind}(M_C)\le n(A)-(d(B)+m)<0\).
From the proof of Lemma 2, we can see: If A is an upper semi-Fredholm operator with \(d(A)=\infty \) and \(n(A)>0\), then there exists a left invertible operator \(C\in B(K,H)\) such that \(M_C\) is upper semi-Fredholm, \(n(M_C)=n(A)>0\) and \(\hbox {ind}(M_C)<0\).
By Lemma 1 and Lemma 2, we obtain the following corollary.
Corollary 1
Suppose that \(B\in B(K)\) and \(A\in B(H)\) are given. If A is an upper semi-Fredholm operator with \(d(A)=\infty \) and \(n(A)+n(B)>0\), then there exists a left invertible operator \(C\in B(K,H)\) such that \(M_C\) is an upper semi-Fredholm operator, \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\).
In Lemma 1 and Lemma 2, \(A\in B(H)\) is upper semi-Fredholm with \(d(A)=\infty \). If \(A\in B(H)\) is a Fredholm operator, we can get that:
Lemma 3
Suppose that \(A\in B(H)\) is a Fredholm operator and \(B\in B(K)\) is an upper semi-Fredholm operator. If \(n(A)+n(B)>0\) and \(\hbox {ind}(A)+\hbox {ind}(B)<0\), then there exists \(C\in B_{l}^{-1}(K, H)\) such that \(M_C\) is an upper semi-Fredholm operator, \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\).
Proof
It is obvious that \(M_C\) is an upper semi-Fredholm operator and \(\hbox {ind}(M_C)=\hbox {ind}(A)+\hbox {ind}(B)<0\) for any \(C\in B(K,H)\).
Case 1. \(n(A)>0\). By \(n(M_C)\ge n(A)>0\) for any \(C\in B(K,H)\), the result is obvious.
Case 2. \(n(A)=0\).
Now, \(n(B)>0\) and \(\hbox {dim}N(B)^{\perp }=\hbox {dim}R(B^*)=\infty \). Take \(e_1\in N(B)\) with \(\Vert e_1\Vert =1\) and then we can extend it into an orthonormal basis of K, written as \(\{e_{n}\}_{n=1}^{\infty }\). Put \(M=\hbox {span}\{e_1\}\). Take \(u_1=Aw_1\in R(A)\) with \(\Vert u_1\Vert =1\), and let \(N=\hbox {span}\{u_1\}\). Then, there are isometric invertible linear operators \(T_1:M^{\perp }\rightarrow H\ominus N\) and \(T_2:M\rightarrow N\) such that \(T_2e_1=u_1=Aw_1\). Define \(C:K\rightarrow H\) by
It is clear that \(C\in B(K,H)\) is invertible and so is left invertible. It follows from \(\hbox {span}\left\{ \left( \begin{array}{ccc} -w_1 \\ e_1 \\ \end{array}\right) \right\} \subseteq N(M_C)\) that \(n(M_C)\ge 1>0\).
According to above lemmas, we will give the necessary and sufficient condition for which \(M_C\) is an upper semi-Fredholm operator, \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\) for some \(C\in B_{l}^{-1}(K, H)\) and the relationship between \(n(M_C)\) and n(A) will be described in section 3. However, we naturally propose such a question:
Given \(A\in B(H)\) and \(B\in B(K)\), what is the necessary and sufficient condition for which \(M_C\) is an upper semi-Fredholm operator, \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\) for all left invertible operators \(C\in B(K, H)\)?
In order to answer the above question, we also need to introduce some lemmas as follows.
Lemma 4
Let \(A\in B(H)\) be a left invertible operator with \(d(A)=\infty \) and \(B\in B(K)\). Then, there exists a left invertible operator \(C\in B(K,H)\) such that \(M_C\) is left invertible.
Proof
Since \(\hbox {dim}R(A)^{\perp }=\hbox {dim}K=\infty \), there exists an isometric invertible linear operator \(T:K\rightarrow R(A)^{\perp }\). Define an operator \(C:K\rightarrow H\) by \(Cy=Ty\) for any \(y\in K\). It is clear that C is left invertible. Similar to the Case 1 of Lemma 2, the desired properties are obvious.
Similar to the condition (1) of sufficiency in the proof of Theorem 2, we have the following lemma.
Lemma 5
[12, Corollary 2.3] Let \(B\in B(K)\) be an upper semi-Fredholm operator and \(A\in B(H)\) be a left invertible operator with \(n(B)\le d(A)\), then there exists a left invertible operator \(C\in B(K,H)\) such that \(M_C\) is left invertible.
Let \(T\in B(K)\). If R(T) is closed, then \(T|_M\) is left invertible for all infinite dimensional closed subspaces \(M\subseteq N(T)^{\perp }\). If R(T) is not closed, we can get:
Lemma 6
Suppose that \(T\in B(K)\) and R(T) is not closed. Then, there exists an infinite dimensional closed subspace \(M\subseteq N(T)^{\perp }\) such that \(T|_M\) is left invertible.
Proof
Since R(T) is not closed, there exists an infinite dimensional closed subspace \(E\subseteq R(T)\). Let \(M=\{x\in N(T)^{\perp }: Tx\in E\}\). It is obvious that \(M\subseteq N(T)^{\perp }\). Then, we prove the following results in turn.
-
(1)
\(\hbox {dim}M=\infty \).
Suppose that \(\{y_n\}_{n=1}^{\infty }\) is an orthonormal basis of E and let \(y_i=Te_i\), where \(e_i\in N(T)^{\perp }\), \(i=1, 2, \ldots \). So, \(\{e_n\}_{n=1}^{\infty }\subseteq M\). From \(T|_{N(T)^{\perp }}\) is injective, we can get that \(\{e_n\}_{n=1}^{\infty }\) is linearly independent. Therefore, \(\hbox {dim}M=\infty \).
-
(2)
M is a closed subspace.
It is clear that M is a subspace. Suppose that \(\{x_n\}\subseteq M\) and \(x_n\rightarrow x_0\), so \(Tx_n\rightarrow Tx_0\) (\(n\rightarrow \infty \)). From \(x_n\in M\) and E is closed, we can get that \(Tx_0\in E\). Since \(N(T)^{\perp }\) is closed, we have \(x_0\in N(T)^{\perp }\). Therefore, \(x_0\in M\) and M is closed.
-
(3)
\(T|_M\) is left invertible.
From \(T|_M\) is injective, we only need \(R(T|_M)\) is closed. Suppose that \(Tx_n\rightarrow y_0\) (\(n\rightarrow \infty \)), where \(\{x_n\}\subseteq M\). It follows that \(\{x_n\}\subseteq N(T)^{\perp }\) and \(Tx_n\in E\). So, \(y_0\in E\subseteq R(T)\). We suppose that \(y_0=Tx_0\), where \(x_0\in N(T)^{\perp }\). So \(x_0\in M\) and \(Tx_n\rightarrow y_0=Tx_0\in R(T|_M)\) as \(n\rightarrow \infty \). Therefore, \(R(T|_M)\) is closed and \(T|_M\) is left invertible.
Lemma 7
Assume \(A\in B(H)\) is an upper semi-Fredholm operator with \(d(A)=\infty \) and \(B\in B(K)\) is not an upper semi-Fredholm operator. Then, there exists a left invertible operator \(C\in B(K,H)\) such that \(M_C\) is not an upper semi-Fredholm operator.
Proof
We will use two cases to complete the proof.
Case 1. \(n(B)=\infty \).
Now, there exists a closed subspace \(M\subseteq R(A)^{\perp }\) such that \(\hbox {dim}N(B)^{\perp }=\hbox {dim}M\). Then, there are two isometric invertible linear operators \(T_1:N(B)^{\perp }\rightarrow M\) and \(T_2:N(B)\rightarrow R(A)\). Define \(C:K\rightarrow H\) by
It is clear that C is left invertible. We suppose that \(\{y_n\}_{n=1}^{\infty }\) is an orthonormal basis of N(B), then there exists \(\{x_n\}_{n=1}^{\infty }\) such that \(T_2y_n=Ax_n\). Hence, \(\left\{ \left( \begin{array}{ccc} -x_n \\ y_n \\ \end{array}\right) \right\} _{n=1}^{\infty }\subseteq N(M_C)\) is linearly independent. So, \(n(M_C)=\infty \) and \(M_C\) is not an upper semi-Fredholm operator.
Case 2. \(n(B)<\infty \) and R(B) is not closed.
Now, \(\hbox {dim}N(B)^{\perp }=\hbox {dim}R(B^*)=\infty \), it follows from Lemma 6 that there exists an infinite dimensional closed subspace \(M\in N(B)^{\perp }\) such that \(B|_M\) is left invertible. Put \(N(B)^{\perp }=M\oplus M^{\perp }\). Since \(R(B)=R(B|_M)+R(B|_{M^{\perp }})\), \(R(B|_M)\) is closed and R(B) is not closed, we can get that \(B|_{M^{\perp }}\) is not left invertible and \(\hbox {dim}M^{\perp }=\infty \). So, there exists \(\{y_n\}_{n=1}^{\infty }\subseteq M^{\perp }\) with \(\Vert y_n\Vert =1\) such that \(By_n\rightarrow 0\) (\(n\rightarrow \infty \)). Let \(T_1:M^{\perp }\rightarrow R(A)\) and \(T_2:M\oplus N(B)\rightarrow R(A)^{\perp }\) be isometric invertible. Define \(C\in B(K, H): K\rightarrow H\) by
It is clear that \(C\in B(K,H)\) is invertible and so is left invertible. From \(C(N(B))\subseteq R(A)^{\perp }\) we obtain that \(N(M_C)=N(A)\oplus \{0\}\). We suppose that \(T_1y_n=Ax_n\), where \(x_n\in N(A)^{\perp }\). Then, \(\left\{ \left( \begin{array}{ccc} -x_n \\ y_n \\ \end{array}\right) \right\} \subseteq N(M_C)^{\perp }\) and \(M_C\left( \begin{array}{c} -x_n \\ y_n \\ \end{array} \right) =\left( \begin{array}{c} 0 \\ By_n \\ \end{array} \right) \rightarrow 0\) (\(n\rightarrow \infty \)). From the fact that \(\Vert Ax_n\Vert =\Vert T_1y_n\Vert =\Vert y_n\Vert =1\), \(x_n\in N(A)^{\perp }\) and \(A|_{N(A)^{\perp }}\) is left invertible, we can get that there is some \( k>0\) such that \(\Vert Ax_n\Vert =1\ge k\Vert x_n\Vert \) for any n, which implies \(\Vert x_n\Vert \le \frac{1}{k}\). So \(1\le \Vert w_n\Vert =\sqrt{\Vert x_n\Vert ^2+\Vert y_n\Vert ^2}\le \sqrt{1+\frac{1}{k^2}}\). Let \(w_n=\left( \begin{array}{c} -x_n \\ y_n \\ \end{array} \right) \) and \(u_n=\frac{w_n}{\Vert w_n\Vert }\), then \(\Vert u_n\Vert =1\) and \(u_n\in N(M_C)^{\perp }\). It follows from \(M_C\left( \begin{array}{c} -x_n \\ y_n \\ \end{array} \right) \rightarrow 0\) that \(M_C(u_n)\rightarrow 0\) (\(n\rightarrow \infty \)) and hence \(M_C|_{N(M_C)^{\perp }}\) is not left invertible. From \(M_C|_{N(M_C)^{\perp }}\) is injective, we get that \(R(M_C|_{N(M_C)^{\perp }})\) is not closed. It follows that \(M_C\) is not an upper semi-Fredholm operator since \(R(M_C)=R(M_C|_{N(M_C)^{\perp }})\).
3 The Proof of the Main Results
Proof of Theorem 1
\(``\Leftarrow "\) From Lemma 3 and Corollary 1, the result is obvious.
\(``\Rightarrow "\) If \(M_C\) is an upper semi-Fredholm operator, then A is an upper semi-Fredholm operator and \(n(A)+n(B)\ge n(M_C)>0\). If \(d(A)=\infty \), then the condition (2) holds. If \(d(A)<\infty \), then A is a Fredholm operator and hence B is an upper semi-Fredholm operator ([13, Theorem 2.2]), \(\hbox {ind}(A)+\hbox {ind}(B)=\hbox {ind}(M_C)<0\). Therefore, the condition (1) holds.
In the proof of Lemma 1 and Lemma 2, we found that there is a connection between \(n(M_C)\) and n(A). Therefore, we give the following corollary.
Corollary 2
Let \(A\in B(H)\) and \(B\in B(K)\). If A is an upper semi-Fredholm operator with \(d(A)=\infty \) and \(n(A)+n(B)>0\), then there exists a left invertible operator \(C\in B(K,H)\) such that \(M_C\) is an upper semi-Fredholm operator, \(\hbox {ind}(M_C)<0\), and \(n(M_C)=\hbox {max}\{n(A), 1\}\).
In Corollary 2, the opposite is not true. For example, let \(A, B\in B(\ell ^2)\) be defined by
It is clear that \(M_C\) is upper semi-Fredholm and \(\hbox {ind}(M_C)=-1<0\) for all \(C\in B(\ell ^2)\). From \(n(A)\le n(M_C)\le n(A)+n(B)\), we can get that \(n(M_C)=n(A)=\hbox {max}\{n(A), 1\}\). But \(d(A)<\infty \). Therefore, Corollary 2 is just a sufficient condition. Theorem 2, as a necessary and sufficient condition for this connection, plays an important role in the proof of Theorem 3. Next, we will give the proof of Theorem 2.
Proof of Theorem 2
\(``\Rightarrow "\) It is obvious that A is an upper semi-Fredholm operator, \(n(A)+n(B)>0\). If \(d(A)=\infty \), then the condition (3) holds. Without loss of generality, we suppose that \(d(A)<\infty \), then A is a Fredholm operator and hence B is an upper semi-Fredholm operator, \(\hbox {ind}(A)+\hbox {ind}(B)=\hbox {ind}(M_C)<0\). We have two cases to complete this proof.
Case 1. \(n(A)>0\).
From \(n(M_C)=\hbox {max}\{n(A), 1\}\), we obtain that \(n(M_C)=n(A)\). By using [13, Corollary 2.6], we can get that \(n(B)\le d(A)\). Therefore, (1) holds.
Case 2. \(n(A)=0\).
From \(n(M_C)=\hbox {max}\{n(A), 1\}\), we obtain that \(n(M_C)=1\). If \(n(B)>d(A)+1\), we suppose that \(n(B)=m\), \(d(A)=k\) and hence \(m-1>k\). Let \(\{e_i\}_{i=1}^{m}\) be an orthonormal basis of N(B) and put \(Ce_i=\alpha _i+\beta _i\), where \(\alpha _i\in R(A)\) and \(\beta _i\in R(A)^{\perp }\), \(i=1,2,\cdots , m\). So, \(\{Ce_1-\alpha _1, \cdots , Ce_{m-1}-\alpha _{m-1}\}\subseteq R(A)^{\perp }\). It follows from \(m-1>k\) that \(\{Ce_1-\alpha _1, \cdots , Ce_{m-1}-\alpha _{m-1}\}\) is a linearly dependent set. Therefore, there exists \(\{a_i\}_{i=1}^{m-1}\subseteq \mathbb {C}\) such that \(a_j\ne 0\) for some j and \(\sum \limits _{i=1}^{m-1}a_iCe_i=\sum \limits _{i=1}^{m-1}a_i\alpha _i\in R(A)\). Put \(Ax_1=\sum \limits _{i=1}^{m-1}a_i\alpha _i\), then we get that \(\left( \begin{array}{c} -x_1 \\ a_1e_1+\cdots +a_{m-1}e_{m-1} \\ \end{array} \right) \in N(M_C)\) and \(\left( \begin{array}{c} -x_1 \\ a_1e_1+\cdots +a_{m-1}e_{m-1} \\ \end{array} \right) \ne 0.\) We may directly assume that \(a_1\ne 0\) for convenience. Similar to the discussion above, we could see that \(\{Ce_2-\alpha _2, \cdots , Ce_{m}-\alpha _{m}\}\) is linearly dependent, so there exists \(\{b_i\}_{i=2}^{m}\subseteq \mathbb {C}\) such that \(b_j\ne 0\) for some j and \(\sum \limits _{i=2}^{m}b_iCe_i=\sum \limits _{i=2}^{m}b_i\alpha _i\in R(A)\). Put \(Ax_2=\sum \limits _{i=2}^{m}b_i\alpha _i\), then \(\left( \begin{array}{c} -x_2 \\ b_2e_2+\cdots +b_{m}e_{m} \\ \end{array} \right) \in N(M_C)\) and \(\left( \begin{array}{c} -x_2 \\ b_2e_2+\cdots +b_{m}e_{m} \\ \end{array} \right) \ne 0\).
We claim that \(\left( \begin{array}{c} -x_1 \\ a_1e_1+\cdots +a_{m-1}e_{m-1} \\ \end{array} \right) \) and \(\left( \begin{array}{c} -x_2 \\ b_2e_2+\cdots +b_{m}e_{m} \\ \end{array} \right) \) are linearly independent.
In fact, we suppose that \(k_1\left( \begin{array}{c} -x_1 \\ a_1e_1+\cdots +a_{m-1}e_{m-1} \\ \end{array} \right) +k_2\left( \begin{array}{c} -x_2 \\ b_2e_2+\cdots +b_{m}e_{m} \\ \end{array} \right) \) \( =0,\) then
Since \(\{e_i\}_{i=1}^{m}\) is linearly independent and \(a_1\ne 0\), we get that \(k_1=0\) and \(k_2b_2=\cdots =k_2b_m=0\). From \(b_j\ne 0\) we can get \(k_2=0\), then the claim is proved. It is in contradiction with \(n(M_C)=1\). Therefore, \(n(B)\le d(A)+1\).
\(``\Leftarrow "\) If the condition (3) holds, from Corollary 2 the result is obvious.
If the condition (1) holds, then \(M_C\) is an upper semi-Fredholm operator with \(\hbox {ind}(M_C)<0\) for all left invertible operators \(C\in B(K,H)\). Since \(n(B)\le d(A)\), there is a closed subspace \(M\subseteq R(A)^{\perp }\) such that \(\hbox {dim}N(B)=\hbox {dim}M\). Therefore, there exists an isometric invertible linear operator \(T_2:N(B)\rightarrow M\). Since A and B are not compact, \(\hbox {dim}N(B)^{\perp }=\hbox {dim}R(B^*)=\infty =\hbox {dim}R(A)\). Then, there exists an isometric invertible linear operator \(T_1:N(B)^{\perp }\rightarrow R(A)\). Define \(C:K\rightarrow H\) by
It is clear that C is left invertible. Similar to the (2) of Case 2 in Lemma 2, \(n(M_C)=n(A)\) is trivial.
If the condition (2) holds, then \(n(B)>0\). We take \(e_1\in N(B)\) with \(\Vert e_1\Vert =1\) and then we can extend it into an orthonormal basis of K, written as \(\{e_{n}\}_{n=1}^{\infty }\). Put \(M=\hbox {span}\{e_1\}\) and take \(u_1=Aw_1\in R(A)\) with \(\Vert u_1\Vert =1\). Let \(N=\hbox {span}\{u_1\}\), then there is an isometric invertible linear operator \(T_2:M\rightarrow N\) such that \(T_2e_1=u_1=Aw_1\). Since \(n(B)\le d(A)+1\), there is a closed subspace \(F\subseteq R(A)^{\perp }\) such that \(\hbox {dim}(N(B)\ominus M)=\hbox {dim}F\). Then there is an isometric invertible linear operator \(T_3:N(B)\ominus M\rightarrow F\). Since \(\hbox {dim}N(B)^{\perp }=\infty =\hbox {dim}(R(A)\ominus N)\), there is an isometric invertible linear operator \(T_1:N(B)^{\perp }\rightarrow R(A)\ominus N\). Define a left invertible operator \(C:K\rightarrow H\) by
Let \(\left( \begin{array}{c} x \\ y \\ \end{array} \right) \in N(M_C)\). Then, \(Ax+Cy=0\) and \(By=0\). Let \(y=k_1e_1+\alpha _1\), where \(k_1\in \mathbb {C}\) and \(\alpha _1\in N(B)\ominus M\). It follows from \(Ax+Cy=Ax+T_2(k_1e_1)+T_3\alpha _1=A(x+k_1w_1)+T_3\alpha _1\) that \(A(x+k_1w_1)=-T_3\alpha _1\in R(A)\cap R(A)^{\perp }=\{0\}\). From \(T_3\) is invertible and \(n(A)=0\), we can get that \(\alpha _1=0\) and \(x=-k_1w_1\). Hence, \(N(M_C)\subseteq \hbox {span}\left\{ \left( \begin{array}{ccc} -w_1 \\ e_1 \\ \end{array}\right) \right\} \). The inclusion “\(\supseteq \)" is obvious. Then \(N(M_C)=\hbox {span}\left\{ \left( \begin{array}{ccc} -w_1 \\ e_1 \\ \end{array}\right) \right\} \). Therefore, \(n(M_C)=1\) and \(n(M_C)=\hbox {max}\{n(A), 1\}\). \(\square \)
Proof of Theorem 3
\(``\Rightarrow "\) It is clear that A is upper semi-Fredholm, then we complete this proof from two steps.
Step 1. \(d(A)<\infty \).
Now, A is a Fredholm operator, B is an upper semi-Fredholm operator, and \(n(A)+n(B)<d(A)+d(B)\). Then, we have two cases:
Case 1. \(n(A)>0\).
It is clear that \(n(M_C)=\hbox {max}\{n(A), 1\}=n(A)\). We claim that \(n(B)=0\). If not, we suppose that \(n(B)=n>0\) and \(\{e_i\}_{i=1}^{n}\) is an orthonormal basis of N(B). From \(\hbox {dim}R(A)=\infty \) we can get that there exists a closed subspace \(M\subseteq R(A)\) such that \(\hbox {dim}N(B)=\hbox {dim}M\). It is easy to get \(\hbox {dim}N(B)^{\perp }=\infty =\hbox {dim}(H\ominus M)\). Therefore, there exist two isometric invertible linear operators \(T_1:N(B)\rightarrow M\) and \(T_2:N(B)^{\perp }\rightarrow H\ominus M\). Define \(C:K\rightarrow H\) by
It is clear that \(C\in B(K,H)\) is invertible and so is left invertible. Let \(T_1e_i=Au_i\), \(i=1, 2, \cdots , m\), then \(\left\{ \left( \begin{array}{ccc} -u_i \\ e_i \\ \end{array}\right) \right\} _{i=1}^{n}\subseteq N(M_C)\) is linearly independent. Supposed that \(\{x_i\}_{i=1}^m\) is an orthonormal basis of N(A) and it follows that \(\left\{ \left( \begin{array}{ccc} x_i \\ 0 \\ \end{array}\right) \right\} _{i=1}^{m}\subseteq N(M_C)\). It is easy to find that \(\left\{ \left( \begin{array}{ccc} x_i \\ 0 \\ \end{array}\right) \right\} _{i=1}^{m}\), \(\left\{ \left( \begin{array}{ccc} -u_i \\ e_i \\ \end{array}\right) \right\} _{i=1}^{n}\) are linearly independent. Therefore, \(n(M_C)\ge n(A)+n(B)>n(A)\), a contradiction with \(n(M_C)=n(A)\). Then, \(n(B)=0\) and the condition (1) holds.
Case 2. \(n(A)=0\).
Now, A is left invertible, \(n(M_C)=1\) and \(n(B)>0\). We claim that \(n(B)=1\). If not, then \(n(B)\ge 2\). Suppose that \(n(B)=k\) and \(\{e_i\}_{i=1}^k\) is an orthonormal basis of N(B). Similar to the proof of Case 1, we can get that \(n(M_C)\ge n(B)\ge 2\), a contradiction. From Lemma 5, we obtain that \(n(B)>d(A)\). Hence, \(d(A)=0\) and A is invertible. The condition (2) holds.
Step 2. \(d(A)=\infty \).
From Lemma 4 we can get that \(n(A)>0\). It follows from \(n(M_C)=\hbox {max}\{n(A), 1\}\) that \(n(M_C)=n(A)\). By Lemma 7 we know that B is an upper semi-Fredholm operator. Similar to the proof of Case 1, we can get that \(n(B)=0\). Therefore, the condition (1) holds.
\(``\Leftarrow "\) If the condition (1) holds, it is obvious that \(M_C\) is an upper semi-Fredholm operator and \(\hbox {ind}(M_C)<0\) for all left invertible operators \(C\in B(K,H)\). From \(0<n(A)\le n(M_C)\le n(A)+n(B)\) we can get that \(n(M_C)=\hbox {max}\{n(A), 1\}\).
If the condition (2) holds, it is clear that \(M_C\) is an upper semi-Fredholm operator, \(n(M_C)\le n(B)=1\) and \(\hbox {ind}(M_C)<0\) for all left invertible operators \(C\in B(K,H)\). If there exists a left invertible operator \(C_0\in B(K,H)\) such that \(n(M_{C_0})=0\), then B is left invertible, since \(M_{C_0}\) is left invertible and A is invertible, which contradicts with \(n(B)=1\). Therefore, \(n(M_C)=1\) for all left invertible operators \(C\in B(K,H)\), which implies \(n(M_C)=\hbox {max}\{n(A), 1\}\). \(\square \)
Based on the above auxiliary results and the proof process of Theorem 3, we can easily obtain the following corollary:
Corollary 3
Let \(A\in B(H)\) and \(B\in B(K)\). Then, \(M_C\) is an upper semi-Fredholm operator, \(n(M_C)>0\) and \(\hbox {ind}(M_C)<0\) for all left invertible operators \(C\in B(K,H)\) if and only if one of the following statements holds:
-
(1)
A and B are both upper semi-Fredholm with \(n(A)>0\) and \(n(A)+n(B)<d(A)+d(B)\);
-
(2)
A is left invertible, B is upper semi-Fredholm with \(n(B)>d(A)\) and \(n(A)+n(B)<d(A)+d(B)\).
References
Aiena, P.: Fredholm and Local Spectral Theory, with Applications to Multipliers. Kluwer Academic Publishers, Dordrecht (2004)
Hai, G.J., Chen, A.: On the right(left) invertible completions for operator matrices. Integr. Equ. Oper. Theory 67(1), 79–93 (2010). https://doi.org/10.1007/s00020-010-1771-1
Dong, J., Cao, X.H.: The semi-Fredholmness and property \((\omega )\) for \(2\times 2\) upper triangular operator matrices. Linear. Multilinear. A. (2023). https://doi.org/10.1080/03081087.2023.2266108
Qiu, S.N., Cao, X.H.: Property (UWE) for operators and operator matrices. J. Math. Anal. Appl. 509(2), 125951 (2022). https://doi.org/10.1016/j.jmaa.2021.125951
Wu, X.F., Huang, J.J.: Weyl spectrum of upper triangular operator matrices. Acta Math. Sin. (Engl. Ser.) 36(7), 783–796 (2020). https://doi.org/10.1007/s10114-020-9485-z
Yang, L.L., Cao, X.H.: Perturbations of spectra and property (R) for upper triangular operator matrices. J. Math. Anal. Appl. 519(1), 126797 (2023). https://doi.org/10.1016/j.jmaa.2005.10.073
Cao, X.H., Meng, B.: Essential approximate point spectra and Weyl’s theorem for operator matrices. J. Math. Anal. Appl. 304(2), 759–771 (2005). https://doi.org/10.1016/j.jmaa.2004.09.053
Wu, X.F., Huang, J.J., Chen, A.: Self-adjoint perturbations of spectra for upper triangular operator matrices. Linear Algebra Appl 531, 1–21 (2017). https://doi.org/10.1016/j.laa.2017.05.017
Wu, X.F., Huang, J.J., Chen, A.: Self-adjoint perturbations of left(right) Weyl spectrum for upper triangular operator matrices. Filomat 36(13), 4385–4395 (2022). https://doi.org/10.2298/FIL2213385W
Dong, J., Cao, X.H.: CFI upper triangular operator matrices. Linear. Multilinear. A. 24, 1–11 (2022). https://doi.org/10.1080/03081087.2022.2058457
Li, Y., Du, H.K.: The intersection of essential approximate point spectra of operator matrices. J. Math. Anal. Appl. 323(2), 1171–1183 (2006). https://doi.org/10.1016/j.jmaa.2015.09.061
Li, Y., Sun, X.H., Du, H.K.: The intersection of left(right) spectra of \(2\times 2\) upper triangular operator matrices. Linear Algebra Appl 418(1), 112–121 (2006). https://doi.org/10.1016/j.laa.2006.01.020
Cao, X.H., Guo, M.Z., Meng, B.: Semi-Fredholm spectrum and Weyl’s theorem for operator matrices. Acta Math. Sin. (Engl. Ser.) 22(1), 169–178 (2006). https://doi.org/10.1007/s10114-004-0505-1
Funding
This work was supported by the National Natural Science Foundation of China (Grant No. 12101081), Fundamental Research Program of Shanxi Province (Grant No. 20210302124079).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declared that they have no conflict of interest.
Additional information
Communicated by Mohammad Sal Moslehian.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Zhang, T., Cao, X. & Dong, J. The Upper Semi-Weylness and Positive Nullity for Operator Matrices. Bull. Malays. Math. Sci. Soc. 47, 80 (2024). https://doi.org/10.1007/s40840-024-01654-y
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s40840-024-01654-y