Abstract
Let \(\mathcal {H}\) be a Hilbert space. The recently introduced notions of the DMP inverse are extended from matrices to operators. The group, Moore–Penrose, Drazin inverses are integrated by DMP inverse and many closely equivalent relations among these inverses are investigated by using appropriate idempotents. Some new properties of DMP inverse are obtained and some known results are generalized.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction and preliminaries
Let \(\mathcal {H}\) be a complex Hilbert space. Denote by \(\mathcal {B(H)}\) of all bounded linear operators on \(\mathcal {H}\). \(\mathcal {R}(T)\) and \(\mathcal {N}(T)\) represent the range and the null space of T, respectively. We call \(P\in \mathcal {B(H)}\) an idempotent if \(P=P^2\), and an orthogonal projector if \(P^2=P=P^*\). The orthogonal projector onto a closed subspace \(\mathcal {M}\) is denoted by \(P_{\mathcal {M}}\). An operator S is an outer generalized inverse of T if \( \mathrm{(II)} \ STS=S. \) Let
The (I, II, III, IV)-inverse is called Moore–Penrose inverse (for short MP inverse), denoted by \(S=T^\dag \). It is well known that T has the MP inverse if and only if \({\mathcal {R}}(T)\) is closed and the MP inverse of T is unique (see [2, 7]). And the (II, V, VI)-inverse is called Drazin inverse, denoted by \(S=T^D\), where \(k=i(T)\) is the Drazin index of T. An operator \(T\in \mathcal {B(H)}\) is Drazin invertible if and only if it has finite ascent asc(T) and descent des(T), which is equivalent with that 0 is a finite order pole of the resolvent operator \(R_\lambda (T)=(\lambda I-T)^{-1}\) [16], say of order k. In such case \(i(T)=asc(T)=des(T)=k\) [2, 4, 16]. Similarly, the (I, II, V)-inverse is called group inverse, denoted by \(S=T^{\#}\) [2, 12]. In the case where \(i(A)\le 1\), \(A^D\) is reduced to the group inverse \(A^{\#}\) [3, 6, 8, 19].
Recently, Baksalary and Trenkler introduced in [1] a new pseudoinverse of a matrix named core inverse. Malik and Thome in [15] generalized this definition and defined a new generalized inverse of a square matrix of an arbitrary index. They used the Drazin inverse (D) and the Moore–Penrose (MP) inverse and therefore this new generalized inverse is called the DMP-inverse (see also [5, 13, 14, 17]).
Definition 1.1
[1, 15] Let closed range operator \(T \in \mathcal {B(H)}\) have index k. Then an operator \(X\in \mathcal {B(H)}\) is the DMP-inverse of T, denoted by \(X=T^{D,\dag }\), if
Our aim is to investigate the characterizations and the properties of DMP inverse. The matrix representation of the DMP inverse is given. We show that all kinds of general inverses and corresponding related idempotents are closed related. Some equivalent characterizations among the existence of these inverses by the existence of self-adjoint idempotents
and idempotents
are built.
2 Some lemmas
To prove the main results, some lemmas are needed.
Lemma 2.1
([7, Theorem 6]) Let \(T_{11}\in \mathcal { B(H)}, T_{22}\in \mathcal {B(K)}, T_{12}\in \mathcal {B(K, H)}\) and \(T_{11}\) be invertible. Then \(T=\left( \begin{array}{c@{\quad }c} T_{11}&{}T_{12}\\ 0&{}T_{22}\end{array}\right) \) is MP invertible if and only if \({\mathcal {R}}(T_{22})\) is closed, and
where \(\triangle =\left[ T_{11}T_{11}^*+T_{12}(I-T_{22}^\dag T_{22})T_{12}^*\right] ^{-1}.\) Moreover, \(TT^\dag =\left( \begin{array}{c@{\quad }c} I&{}0\\ 0&{}T_{22}T_{22}^\dag \end{array}\right) \) and
If T has the Drazin inverse \(T^D\) with \(i(T)=k\), then \({\mathcal {R}}(T^k) \) is an invariant subspace of T since \(TT^k=T^{k+2}T^D=T^kT^2T^D.\) And T has the following operator matrix
with respect to the space decomposition \({\mathcal {H}}={\mathcal {R}}(T^k)\oplus {\mathcal {R}}(T^k)^\perp ,\) where \(T_{11}\) is invertible and \(T_{22}^k=0\) [10].
Lemma 2.2
([10, Theorem 2.5]) If \(T\in \mathcal {B(H)}\) is Drazin invertible with \(i(T)=k\), then T has the operator matrix form (4) and
Throughout this work we denote by
Lemma 2.3
Let \(X_0\) be defined as in (6), where \(T_{11}\) is invertible and \(T_{22}^k=0.\) Then
-
(i)
$$\begin{aligned} T_{11}X_0-X_0T_{22}=T_{11}^{-1}T_{12}. \end{aligned}$$(7)
-
(ii)
\(X_0=0\Longleftrightarrow T_{12}=0.\)
-
(iii)
\(X_0T_{22}T_{22}^\dag =0\Longleftrightarrow X_0T_{22}=0\Longleftrightarrow T_{12}T_{22}=0\Longleftrightarrow X_0=T^{-2}_{11}T_{12}.\)
Proof
Item (i) is clear by the definition of \(X_0\) in (6). Item (ii) follows by the relation in (7).
(iii) If \(X_0T_{22}T_{22}^\dag =0\), then \(X_0T_{22}=X_0T_{22}T_{22}^\dag T_{22}=0\).
If \(X_0T_{22}=0\), by (6),
Product \(T^{k-2}_{22}\) from right in (8) we get \(T^{-2}_{11}T_{12}T^{k-1}_{22}=0\). It follows that \(T_{12}T^{k-1}_{22}=0\). In the same way, \(T_{12}T^{k-2}_{22}=0\) by production \(T^{k-3}_{22}\) from right in (8). With a step by step deduction it follows that \(T_{12}T_{22}=0\).
If \(T_{12}T_{22}=0\). then \(X_0=T^{-2}_{11}T_{12}\) by the definition of \(X_0\).
On the other hand, if \(X_0=T^{-2}_{11}T_{12},\) then
Again we derive that \(T_{12}T_{22}=0\) by the above method.
If \(T_{12}T_{22}=0\), then \(X_0T_{22}=0\). If \(X_0T_{22}=0 \), it is clear that \( X_0T_{22}T_{22}^\dag =0\). \(\square \)
The next well-known criterion (i) due to Douglas [9] (see also Fillmore and Williams [11]) about range inclusions and factorization of operators will be crucial. The criterion (ii) was given in [18] for Hilbert C*-modules.
Lemma 2.4
-
(i)
[9] If \(A,B\in \mathcal {B(H)}\), there exists an operator \(C\in \mathcal {B(H)}\) such that \(A=BC\) if and only if \(\mathcal {R}(A)\subset \mathcal {R}(B)\).
-
(ii)
[18] If \(\mathcal {L}\) and \(\mathcal {M}\) are closed subspaces of \(\mathcal {H}\) and \(P_{{\mathcal {L},\mathcal {M}}}\) is an idempotent on \(\mathcal {L}\) along \(\mathcal {M}\), then
$$\begin{aligned} P_{{\mathcal {L},\mathcal {M}}}T=T \Longleftrightarrow {\mathcal {R}}(T)\subset \mathcal {L},\quad TP_{{\mathcal {L},\mathcal {M}}}=T\Longleftrightarrow {\mathcal {N}}(T)\supset \mathcal {M}. \end{aligned}$$
3 The representation for the DMP-inverse
First we show that the solution of (1) is unique if system is consistent (see [15, Definition2.3] for the matrix case).
Theorem 3.1
If closed range operator \(T \in \mathcal {B(H)}\) has index k, then the DMP inverse of T is unique and \(T^{D,\dag }=T^DTT^\dag .\)
Proof
T is MP invertible and Drazin invertible since T is closed range operator with index k (not necessarily \(\le 1\)). Let \(X=T^DTT^\dag \). Then
\(XT=T^DTT^\dag T=T^DT\) and \(T^kX=T^kT^DTT^\dag =T^kT^\dag \). Hence \(X=T^DTT^\dag \) is a solution of system (1). If \(X_1\) and \(X_2\) are two solutions of system (1), then
Hence system (1) has a unique solution. \(\square \)
From Theorems 3.1 it follows that T is both Drazin and MP invertible if and only if T is DMP invertible. We next give the canonical form for the DMP inverse of an operator T using block operator matrix method.
Theorem 3.2
Let closed range operator \(T \in \mathcal {B(H)}\) have the operator matrix form (4) and \(i(T)=k\). Then
with respect to the space decomposition \({\mathcal {H}}={\mathcal {R}}(T^k)\oplus {\mathcal {R}}(T^k)^\perp ,\) where \(X_0\) is defined in (6).
Proof
By Theorem 3.1, Lemmas 2.1 and 2.2,
There is another inverse associated with operator T, namely \(T^{\dag , D}=T^\dag TT^D\) and its canonical form in terms of the decomposition of T in (4) is given by
where \(X_0\) and \(\triangle \) are defined in (6).
Remark
(1) If \((I-T_{22}^\dag T_{22}) T_{12}^*=0\), then \(T^{\dag , D}=T^{D}.\) If \(i(T)\le 1\), then \(T_{22}=0\). Hence, \(T^{\#,\dag }=T_{11}^{-1}\oplus 0\),
where \(\triangle '=(T_{11}T_{11}^*+T_{12}T_{12}^*)^{-1}.\)
(2) If \(T^D=T^\dag \), then \(T_{12}=0\) and \(T_{22}=0\) in (4). Hence
(3) If \(T^D=T\), then \(T_{22}=0\) and \(T^2_{11}=I\) in (4). Hence, \(i(T)\le 1\) and \(T=T^3\),
and
(4) If \(T^{D, \dag }=T\), then \(T_{12}=0\), \(T_{22}=0\) and \(T^2_{11}=I\) in (4). If \(T^\dag =T\), by (2) and (4), \((I-T_{22}^\dag T_{22})T_{12}^*=0\) and \(T^\dag _{22}=T_{22}\). Since \(T_{22}\) is k nilpotent, \(T^\dag _{22}=T_{22}\) implies that \(T_{22}=0\). It follows that \(T_{12}=0\) and \(T^2_{11}=I\). Then we derive that
In this case,
In the following we introduce a method to obtain the DMP inverse by a different algebraic approach.
Theorem 3.3
Let closed range operator \(T \in \mathcal {B(H)}\) have index k. Then \(T^{D,\dag }=(T^2T^\dag )^D\).
Proof
Let \(T\in \mathcal {B(H)}\) have the operator matrix form (4) and \(i(T)=k\). By Theorem 3.2,
By Lemma 2.1,
If \(i(T)\le 1\), then \(T_{22}=0\) and
The result holds. If \(k\ge 2\), then \((T^2_{22}T_{22}^\dag )^{k-1}=T^{k}_{22}T_{22}^\dag =0\) since \(T_{22}^k=0\). By Lemma 2.2,
where
Hence, \(T^{D,\dag }=(T^2T^\dag )^D\). \(\square \)
4 Characterizations of relating idempotents
Suppose that \(T \in \mathcal {B(H)}\) is a closed range operator with index k. Then \(\mathcal {R}(T^\dag )=\mathcal {R}(T^*)\) and \(\mathcal {N}(T^\dag )=\mathcal {N}(T^*)\). By Lemma 2.4,
and
The outer generalized inverse \(T^{(2)}_{\mathcal {S},\mathcal {T}}\) of \(T\in \mathcal {B(H)}\) is the operator \(X\in \mathcal {B(H)}\) satisfying \(XTX=X\), \(\mathcal {R}(X)=\mathcal {S}\) and \(\mathcal {N}(X)=\mathcal {T}\) [2, 4]. Hence, \(T^\dag \), \(T^D\) and \(T^{\#}\) are outer generalized inverse \(T^{(2)}_{\mathcal {S},\mathcal {T}}\) with prescribed range \(\mathcal {S}\) and null space \(\mathcal {T}\):
The DMP-inverse is one kind of outer generalized inverses \( T^{D,\dag }=T^{(2)}_{\mathcal {R}(T^k), \mathcal {N}(T^DTT^\dag )}.\) If \(i(T)\le 1,\) then
and \( T^{\#,\dag }=T^{(2)}_{\mathcal {R}(T), \mathcal {N}(T^*)}.\) Since \(T^{D,\dag }=T^DTT^\dag \) and \(T^{\dag ,D}=T^\dag TT^D,\)
Hence \(T^{\dag , D} T\), \(TT^{\dag , D} \), \(TT^{D, \dag } \) and \(T^{D, \dag } T\) are idempotents. Define the self-adjoint idempotents
and the idempotents
We have the following equivalent relations.
Theorem 4.1
Let closed range operator \(T \in \mathcal {B(H)}\) have index k. Let \(Q_i, i=1,2,3\), be defined as in rm (13). Then
-
(i)
\(Q_1=Q_2\Longleftrightarrow \mathcal {N}(T^*)\subseteq \mathcal {N}(T^k)\Longleftrightarrow T^D=T^{D,\dag }.\)
-
(ii)
\(Q_1=Q_3\Longleftrightarrow \mathcal {R}(T^k)\subseteq \mathcal {R}(T^*)\Longleftrightarrow T^D=T^{\dag ,D}.\)
-
(iii)
\(Q_2=Q_3\Longleftrightarrow \mathcal {N}(T^*)\subseteq \mathcal {N}(T^k)\) and \(\mathcal {R}(T^k)\subseteq \mathcal {R}(T^*) \Longleftrightarrow T^{D,\dag }=T^D=T^{\dag ,D}.\)
Proof
(i) Since \(Q_1=T^{D,\dag } T=TT^D\) and \(Q_2=TT^{D,\dag }=TT^DTT^\dag \),
(ii) Since \(Q_3=T^{\dag , D}T=T^\dag TT^D T\),
(iii) We only show that \(Q_2=Q_3\Longrightarrow \mathcal {N}(T^*)\subseteq \mathcal {N}(T^k)\) and \(\mathcal {R}(T^k)\subseteq \mathcal {R}(T^*).\) The rest results are obvious by items (i) and (ii). If \(Q_2=Q_3\), then \(TT^DTT^\dag =T^\dag TT^DT\). First, product T from left we get \(T^2T^DTT^\dag =T^2T^D\). So \(T^2T^D(I-TT^\dag )=0\), which implies that \(\mathcal {N}(T^*)=\mathcal {N}(TT^\dag )=\mathcal {R}(I-TT^\dag )\subseteq \mathcal {N}(T^2T^D)=\mathcal {N}(T^D)=\mathcal {N}(T^k)\). Second, product T from right we get \(T^DT^2=T^\dag TT^DT^2\). So \((I-T^\dag T)T^2T^D=0\), which implies that \(\mathcal {R}(T^k)=\mathcal {R}(T^2T^D)\subseteq \mathcal {N}(I-T^\dag T)=\mathcal {R}(T^*)\). \(\square \)
Theorem 4.2
Let closed range operator \(T \in \mathcal {B(H)}\) have index k. Let \(P_i\) and \(Q_i, i=1,2,3\), be defined as in (12) and (13), respectively. Then
-
(i)
$$\begin{aligned} \begin{array}{rcl} P_2=Q_1&{} \Longleftrightarrow &{} P_2=Q_3\Longleftrightarrow [T, P_2]=:TP_2-P_2T=0\\ &{}\Longleftrightarrow &{} T_{12}=0. \quad (\hbox {see }\ (4) \ \hbox {for }\ T_{i2}, i=1,2)\end{array} \end{aligned}$$
-
(ii)
\(P_2=Q_2\Longleftrightarrow P_2T(I-P_2)T=0\Longleftrightarrow T_{12}T_{22}=0\).
-
(iii)
\(P_1=P_2\Longleftrightarrow P_1=Q_2\Longleftrightarrow i(T)\le 1\) (i.e., T is group invertible)\(\Longleftrightarrow T_{22}=0\).
-
(iv)
$$\begin{aligned} \begin{array}{rcl} P_1= P_3&{} \Longleftrightarrow &{} P_2=P_3\Longleftrightarrow P_1= Q_1 \Longleftrightarrow P_1=Q_3\\ &{}\Longleftrightarrow &{} P_3= Q_1 \Longleftrightarrow P_3=Q_2 \Longleftrightarrow P_3=Q_3 \\ &{}\Longleftrightarrow &{} i(T)\le 1\quad {\text{ and }} \quad \mathcal {R}(T)=\mathcal {R}(T^*)\quad (i.e., T \ \hbox {is EP}) \\ &{}\Longleftrightarrow &{} T_{12}=0 \quad {\text{ and }} \quad T_{22}=0.\end{array} \end{aligned}$$
Proof
We give the proof only for the items (i) and (ii); the other items may be proved in the same manner.
and
Then \(P_2=Q_1 \Longleftrightarrow X_0=0\Longleftrightarrow T_{12}=0\) by Lemma 2.3 \(\Longleftrightarrow P_2=Q_3\Longleftrightarrow [T, P_2]=:TP_2-P_2T=0\).
Then \(P_2=Q_2\Longleftrightarrow X_0T_{22}T_{22}^\dag =0 \Longleftrightarrow T_{12}T_{22}=0\) by Lemma 2.3 \(\Longleftrightarrow P_2T(I-P_2)T=0\). \(\square \)
Theorem 4.3
Let closed range operator \(T \in \mathcal {B(H)}\) have index k. Let \(P_i\) and \(Q_i, i=1,2,3\), be defined as in (12) and (13), respectively. Then
-
(i)
\(P_1T^D=T^DP_3=T^D\).
-
(ii)
\(P_1Q_1=Q_1=Q_1P_3\) and \(P_2P_1=P_2=P_1P_2.\)
-
(iii)
\(T^{D, \dag }=T^DP_1=Q_1T^DP_1\) and \(T^{\dag , D}=P_3T^D=P_3T^DQ_1.\)
-
(vi)
\(Q_1TP_3=P_1TQ_1=Q_1TQ_1=TQ_1=T^2T^D=(T^D)^D.\)
-
(v)
\(T^2\left[ T^{D,\dag }\right] ^2= TT^{D,\dag }=Q_2\) and \(\left[ T^{D,\dag }\right] ^2T^2=T^{D,\dag }T= T^DT=Q_1.\)
Proof
We only give the proof of the items (i) and (v). The other items can be checked by the definitions in (12) and (13).
(i) Since \({\mathcal {R}}(T^D)={\mathcal {R}}(T^k)\subseteq {\mathcal {R}}(T)\) and \({\mathcal {N}}(T^D)={\mathcal {N}}(T^k)\supseteq {\mathcal {N}}(T),\) by Lemma 2.4,
(iv) \(T^2\left[ T^{D,\dag }\right] ^2=T^2 T^DTT^\dag TT^DT^\dag =T^2T^DTT^DT^\dag =T^2T^DT^\dag = TT^{D,\dag }=Q_2\) and \(\left[ T^{D,\dag }\right] ^2T^2= T^DTT^\dag T^DTT^\dag T^2=T^DT=T^DTT^\dag T=T^{D,\dag }T=Q_1.\) \(\square \)
It is well known that \((T^D)^D =T^2T^D\) and \((T^\dag )^\dag =T\). The relations for \((T^{D, \dag })^{D,\dag }\), \((T^{D})^{D,\dag }\) and \((T^{D,\dag })^{D}\) are given as follows.
Theorem 4.4
Let closed range operator \(T \in \mathcal {B(H)}\) have index k. Let \(P_2\) be defined as in (12). Then
-
(i)
\((T^{D, \dag })^{D,\dag }= (T^{D})^{D,\dag }=TP_2\).
-
(ii)
\(\left[ (T^{D})^{D,\dag }\right] ^2 T^{D,\dag }=T^2T^{D,\dag }= (T^{D,\dag })^D.\)
Proof
(i) Let \(T\in \mathcal {B(H)}\) have the operator matrix form (4) and \(i(T)=k\). By Lemma 2.2 and Theorem 3.2,
where \(\Delta ''=\left[ T_{11}^{-1}(T_{11}^{-1})^* +X_0T_{22}T_{22}^\dag (X_0T_{22}T_{22}^\dag )^*\right] ^{-1}.\) So,
In the same vein, we obtain that
Hence (i) holds.
(ii) By the proof of item (i),
and
\(\square \)
References
Baksalary, O.M., Trenkler, G.: Core inverse of matrices. Linear Multilinear Algebra 58(6), 681–697 (2010)
Ben-Israel, A., Greville, T.N.E.: Generalized Inverses: Theory and Applications, 2nd edn. Springer, New York (2003)
Bouldin, R.H.: Generalized inverses and factorizations, recent applications of generalized inverses. Pitman Ser. Res. Notes Math. 66, 233–249 (1982)
Campbell, S.L., Meyer, C.D.: Generalized Inverses of Linear Transformations. Dover, New York (1991)
Coll, C., Thome, N.: Oblique projectors and group involutory matrices. Appl. Math. Comput. 140, 517–522 (2003)
Cvetković-Ilić, D.S.: A note on the representation for the Drazin inverse of \(2 \times 2\) block matrices. Linear Algebra Appl. 429, 242–248 (2008)
Deng, C., Du, H.: Representations of the Moore–Penrose inverse for a class of \(2\times 2\) block operator valued partial matrices. Linear Multilinear Algebra 58(1), 15–26 (2009)
Djordjević, D.S., Wei, Y.: Additive results for the generalized Drazin inverse. J. Aust. Math. Soc. 73, 115–125 (2002)
Douglas, R.G.: On majorization, factorization and range inclusion of operators in Hilbert space. Proc. Am. Math. Soc. 17, 413–416 (1966)
Du, H., Deng, C.: The representation and characterization of Drazin inverses of operators on a Hilbert space. Linear Algebra Appl. 407, 117–124 (2005)
Fillmore, P.A., Williams, J.P.: On operator ranges. Adv. Math. 7, 254–281 (1971)
Hartwig, R.E., Shoaf, J.M.: Group inverses and Drazin inverses of bidiagonal and triangular Toeplitz matrices. Aust. J. Math. 24A, 10–34 (1977)
Herrero, A., Ramírez, F., Thome, N.: Relationships between different sets involving group and Drazin projectors and non-negativity. Linear Algebra Appl. 438, 1688–1699 (2013)
Lebtahi, L., Thome, N.: The spectral characterization of k-generalized projectors. Linear Algebra Appl. 420, 572–575 (2007)
Malik, S.B., Thome, Néstor: On a new generalized inverse for matrices of an arbitrary index. Appl. Math. Comput. 226, 575–580 (2014)
Taylar, A.E., Lay, D.C.: Introduction to Functional Analysis, 2nd edn. Wiley, New York (1980)
Thome, N., Wei, Y.: Generalized inverses and a block-rank equation. Appl. Math. Comput. 141, 471–476 (2003)
Xu, Q.: Common hermitian and positive solutions to the adjointable operator equations \(AX=C\), \(XB=D\). Linear Algebra Appl. 429, 1–11 (2008)
Wei, Y., Qiao, S.: The representation and approximation of the Drazin inverse of a linear operator in Hilbert space. Appl. Math. Comput. 138, 77–89 (2003)
Author information
Authors and Affiliations
Corresponding author
Additional information
Partial support was provided by the National Natural Science Foundation of China under Grant 11171222.
Rights and permissions
About this article
Cite this article
Yu, A., Deng, C. Characterizations of DMP inverse in a Hilbert space. Calcolo 53, 331–341 (2016). https://doi.org/10.1007/s10092-015-0151-2
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10092-015-0151-2