Full Text
UDC 517.43 + 517.94
MATHEMATICS
F. S. ROFE-BEKETOV
SELF-ADJOINT EXTENSIONS OF DIFFERENTIAL OPERATORS IN A SPACE OF VECTOR FUNCTIONS
(Presented by Academician A. A. Dorodnitsyn on 10 VI 1968)
-
In the present note we establish the general form of self-adjoint boundary-value problems on the finite interval \([0,b]\) for differential equations \(l[y]=\lambda y\) of arbitrary order \(m\) with continuous operator coefficients. For a scalar quasi-differential expression of even order with real coefficients, a description of all self-adjoint extensions on the interval \([0,b]\) was given in the work of M. G. Krein \((^1)\) and is presented in the monographs \((^2,^3)\). For the algebraic study of scalar boundary-value problems (Bôcher, etc.) see \((^4)\).
-
The basis of the proposed investigation is the concept, introduced by us, of a Hermitian relation.
Definition. A binary relation \(\theta\) given in some Hilbert space \(H\) is called Hermitian if from \(x\theta x'\), \(y\theta y'\), where \(x,x',y,y'\in H\), it follows that
\[ (x',y)-(x,y')=0, \tag{1} \]
and from the validity of (1) for certain \(x,x'\in H\) with all pairs \(y\theta y'\) it follows that also \(x\theta x'\).
For each pair \(x\theta x'\) construct the vectors \(x^\pm=x'\pm ix\). By the equality \(U_\theta x^+=x^-\) a unitary operator \(U_\theta\) is defined, which we shall call the Cayley transform of the Hermitian relation \(\theta\).
Theorem 1. Whatever the self-adjoint operator \(A\) and the unitary operator \(U\), the relation defined by either of the equations
\[ \cos A\cdot x' - \sin A\cdot x = 0, \tag{2} \]
\[ (U-I)x' + i(U+I)x = 0, \tag{3} \]
is Hermitian. Conversely, every Hermitian relation \(x\theta x'\) is representable in the forms (2) and (3), where the unitary operators \(-e^{2iA}\) and \(U\) determine the relation \(\theta\) uniquely and are its Cayley transform \(U_\theta\).
Corollary. Any Hermitian relation \(x\theta x'\) can be represented in the form \(x'=A_1x+x^\perp\), where \(x\in H_1\), \(H_1\) is some subspace of \(H\), \(x^\perp\in H_1^\perp\), and \(A_1\) is a self-adjoint operator in \(H_1\), possibly unbounded*.
Theorem 2. Let \(B\) and \(C\) be arbitrary bounded operators on all of \(H\). The relation \(\theta\) defined by them,
\[ x\theta x' \leftrightarrow Cx' - Bx = 0 \tag{4} \]
is Hermitian if and only if the operators \(B\pm iC\) are invertible on their ranges and the operator \(U=(B+iC)^{-1}(B-iC)\) is unitary. Under these conditions the relation \(\theta\) can be represented through the operator \(U\) by formula (3).
* Thus, the known general form of self-adjoint boundary conditions for elliptic differential equations \((^5)\) also admits interpretation from the point of view of the concept of Hermitian relations.
Corollary. If \(\dim H<\infty\), relation (4) is Hermitian if and only if \(BC^*=CB^*\) and \(\det(BB^*+CC^*)\ne0\).
We shall say that a relation \(\theta\) is the closure of \(\theta\) if the graph \(\widetilde{\theta}\) in \(H\oplus H\) is the closure of the graph of \(\theta\).
Theorem 3. Let the operators \(B\) and \(C\) in (4) be arbitrary (possibly unbounded, nonclosed, and with domains \(D_B\) and \(D_C\) not dense in \(H\)). In order that the closure \(\widetilde{\theta}\) of the relation \(\theta(4)\) be Hermitian, the following conditions are necessary: 1. The linear span \(D_B\cup D_C\) is dense in \(H\). 2. The operators \(B\pm iC\) are invertible on their ranges. 3. The operator
\[
U_1=(B+iC)^{-1}(B-iC)
\]
in \(H_1=D_B\cap D_C\) has a closure \(\overline{U}_1\), which is unitary. 4. \(B\{D_B\cap D_C^\perp\}=\{0\}\), \(C\{D_C\cap D_B^\perp\}=\{0\}\). Conditions 1–4, together with condition 5.
\[
\overline{D_B\cap D_C}=\overline{D_B}\cap\overline{D_C},
\]
become sufficient for the Hermiticity of \(\widetilde{\theta}\), and the Cayley transform of the relation \(\widetilde{\theta}\) is then given by the formula
\[
U_{\widetilde{\theta}}=\overline{U}_1\oplus I_{D_B^\perp}\oplus(-I_{D_C^\perp}).
\]
3. Let \(\mathcal H(0,b)\) be the Hilbert space of vector-functions with values in the separable Hilbert space \(H\) and with scalar product
\[
\langle x,y\rangle=\int_0^b (x(t),y(t))_H\,dt .
\]
Consider in \(\mathcal H(0,b)\) a differential operation \(l[y]\) of order \(m\). For \(m=2n\) put
\[
l[y]=\sum_{k=1}^{n}(-1)^k\{(p_{n-k}y^{(k)})^{(k)}
-i[(q_{n-k}y^{(k)})^{(k-1)}+(q_{n-k}y^{(k-1)})^{(k)}]\}+p_ny.
\tag{5}
\]
Let all operator coefficients be self-adjoint:
\[
p_k(t)=p_k^*(t),\qquad q_k(t)=q_k^*(t)
\tag{6}
\]
and depend continuously on \(t\), together with their derivatives up to order \(n-k\) inclusive, and suppose that \(p_0^{-1}(t)\) exists and is bounded for \(t\in[0,b]\). Define for the operation (5) the quasiderivatives \(y^{[k]}\) by the formulas*
\[
y^{[j]}=y^{(j)}\quad (j=0,1,\ldots,n-1);\qquad
y^{[n]}=p_0y^{(n)}-iq_0y^{(n-1)},
\]
\[
y^{[n+k]}=-\frac{d}{dt}y^{[n+k-1]}+p_ky^{(n-k)}
+i\,[q_{k-1}y^{(n-k+1)}-q_ky^{(n-k-1)}]
\]
\[
(k=1,\ldots,n;\quad q_n\equiv0;\quad l[y]\equiv y^{[2n]}).
\]
Let \(L\) be the operator generated by the expression \(l[y]\) on the set \(D\) of all such \(y(t)\in\mathcal H(0,b)\) with \(m-1\) absolutely continuous (quasi-)derivatives for which \(l[y]\in\mathcal H(0,b)\), and let \(L_0\) be the restriction of \(L\) defined by the conditions
\[
y_0=y_0^{[1]}=\cdots=y_0^{[m-1]}=0,\qquad
y_b=y_b^{[1]}=\cdots=y_b^{[m-1]}=0,
\tag{7}
\]
where \(y_0^{[k]}=y^{[k]}(0)\), \(y_b^{[k]}=y^{[k]}(b)\), \(k=0,1,\ldots\).
Denote
\[
H^m=H\oplus\cdots\oplus H
\]
(\(m\) summands), and let \((\cdot,\cdot)_m\) be the scalar product in \(H^m\). To each vector-function \(u(t)\in D\) we associate a pair of vectors \(\hat u,\hat u'\in H^m\) (for \(m=2n\)):
\[
\hat u=\{u_0,u'_0,\ldots,u_0^{(n-1)},u_b,u'_b,\ldots,u_b^{(n-1)}\},
\]
\[
\hat u'=\{u_0^{[2n-1]},u_0^{[2n-2]},\ldots,u_0^{[n]},-u_b^{[2n-1]},-u_b^{[2n-2]},\ldots,-u_b^{[n]}\}.
\tag{8}
\]
* These requirements are easily weakened by considering \(l\) as a quasidifferential operation.
** For \(q_0=q_1=\cdots=q_{n-1}=0\) our definition coincides with that adopted in \((^{1-3})\).
Then for operation (5) Lagrange’s identity is written in the form
\[ \langle l[x],y\rangle-\langle x,l[y]\rangle=(\hat{x}',\hat{y})_m-(\hat{x},\hat{y}')_m . \tag{9} \]
Lemma 1. The equality \(L_0^*=L\) is valid.
An essential point in the proof of Lemma 1 is
Lemma 2. The manifold of vector solutions \(y=y(t)\) of the homogeneous equation of arbitrary order \(m\):
\[
y^{(m)}+g_1(t)y^{(m-1)}+\cdots+g_m(t)y=0
\]
with arbitrary continuous operator coefficients \(g_k(t)\) forms a subspace in \(\mathcal H(0,b)\) (i.e., a closed one), since the dependence between the solutions \(y(t)\in\mathcal H(0,b)\) and their Cauchy data is mutually continuous.
From (9), Theorem 1, and Lemma 1 there follows
Theorem 4. Between the self-adjoint extensions \(\widetilde L\) of the operator \(L_0\) (5), (7) and the Hermitian relations \(\theta\) in \(H^{2n}\) there exists a one-to-one correspondence, by virtue of which every \(\widetilde L\) is generated by the operation \(l[y]\) (5) and boundary conditions of any of the forms
\[ \cos \hat A\cdot \hat y' - \sin \hat A\cdot \hat y=0, \tag{10} \]
\[ (\hat U-\hat I)\hat y' + i(\hat U+\hat I)\hat y=0, \tag{11} \]
where \(\hat A,\hat U\) are respectively a self-adjoint and a unitary operator in \(H^{2n}\), and \(\hat y,\hat y'\in H^{2n}\) are determined from \(y(t)\in D\) by formula (8). Conversely, any of these boundary conditions determines some self-adjoint extension \(\widetilde L\) of the operator \(L_0\).
Example 1. Separated self-adjoint boundary conditions for \(t=0\) are always reduced to the form*
\[ \cos \hat A_0\cdot \hat y_0' - \sin \hat A_0\cdot \hat y_0=0, \tag{12} \]
where \(\hat A_0\) is an arbitrary self-adjoint operator in \(H^n\),
\[
\hat y_0=\{y_0,y_0,\ldots,y_0^{(n-1)}\},\qquad
\hat y_0'=\{y_0^{[2n-1]},y_0^{[2n-2]},\ldots,y_0^{[n]}\}.
\]
Example 2. Generalized periodic conditions:
\[
y_b^{[2n-k]}=U_k y_0^{[2n-k]},\qquad
y_b^{(k-1)}=U_k y_0^{(k-1)},\quad k=1,\ldots,n;
\]
\(U_k\) are unitary operators in \(H\).
4. Let now \(l\) be an operation of odd order \(m=2n+1\)
\[ l[y]=\sum_{k=0}^{n}(-1)^k\left\{\,i\left[(q_{n-k}y^{(k)})^{(k+1)}+(q_{n-k}y^{(k+1)})^{(k)}\right]+(p_{n-k}y^{(k)})^{(k)}\right\}, \tag{13} \]
where \(q_0^{-1}(t)\) exists and is continuous for \(t\in[0,b]\), and (6) is satisfied. For operation (13) we define quasiderivatives
\[ y^{[j]}=y^{(j)}\quad (j=0,1,\ldots,n-1),\qquad y^{[n]}=-iq_0y^{(n)}, \]
\[ y^{[n+k+1]}=-\frac{d}{dt}y^{[n+k]}+p_k y^{(n-k)} +i\,[q_k y^{(n-k+1)}-q_{k+1}y^{(n-k-1)}] \]
\[ (k=0,1,\ldots,n;\ q_{n+1}\equiv0;\ l[y]\equiv y^{[2n+1]}). \]
Let \(H_t^\pm\) be invariant subspaces of the operator \(q_0(t)\) such that \(q_0(t)>0\) on \(H_t^+\) and \(q_0(t)<0\) on \(H_t^-\), and let \(P_t^\pm\) be the orthoprojectors onto \(H_t^\pm\), respectively. Put
\[
q_\pm(t)=\pm q_0(t)P_t^\pm,\qquad
Q_t^\pm=q_+^{1/2}(t)\pm q_-^{1/2}(t).
\]
From the properties of \(q_0(t)\) it follows that \(\dim H_b^\pm=\dim H_0^\pm\). Let \(U_q\) be an arbitrary but fixed unitary operator in \(H\) carrying \(H_b^\pm\) into \(H_0^\pm\). To each \(v(t)\in D\) (see item 3 for \(m=2n+1\)) we associate a pair \(\hat v,\hat v'\in H^{2n+1}\):
\[ \hat v=\{Q_0^+v_0^{(n)}+U_qQ_b^+v_b^{(n)},\ v_0,\ v_0',\ldots,v_0^{(n-1)},\ v_b,\ v_b',\ldots,v_b^{(n-1)}\}, \tag{14} \]
\[ \text{* For a scalar real equation of even order, conditions of the form (12) are given as self-adjoint in }{}^{(6)}. \text{ For Schrödinger’s operator potential equation } (*)\ -y''+q(t)y=\lambda y,\text{ condition (12) is used in }{}^{(7)}. \text{ The general form of self-adjoint separated boundary conditions for equation }(*)\text{ was obtained by another method and in another form in }{}^{(8)}. \]
\[ \hat v'=\{iU_qQ_bv_b^{(n)}-iQ_0^-v_0^{(n)},\ v_0^{[2n]},\ v_0^{[2n-1]},\ldots,\ v_0^{[n+1]},\ -v_b^{[2n]},\ -v_b^{[2n-1]},\ldots, \ldots,\ -v_b^{[n+1]}\}. \tag{14} \]
Lemma 3. For the operator \(l\) (13), Lagrange’s identity has the form (9), where \(\hat x,\hat y,\hat x',\hat y'\) are defined by formulas (14).
Theorem 5. For the operator \(L_0\) (13), (7) of odd order, Theorem 4 is valid with the replacement in its formulation of \(l[y]\) (5) by \(l[y]\) (13), \(H^{2n}\) by \(H^{2n+1}\), and formulas (8) by (14).
Theorem 6. For the existence of separated self-adjoint boundary conditions for the operation \(l[y]\) (13) of order \(2n+1\) in the cases \(\dim H<\infty\) or \(\dim H=\infty\), but \(n=0\), it is necessary and sufficient that \(\dim H_0^+=\dim H_0^-\le\infty\). In this case all such boundary conditions are representable (for \(t=0\)) in the form (12), where \(\hat A_0\) is a self-adjoint operator in \(H^{n+}=H^n\oplus H_0^+\),
\[ \hat y_0=\{(q_+^{1/2}(0)+V_0q_-^{1/2}(0))y_0^{(n)},\ y_0,\ y_0',\ldots,\ y_0^{(n-1)}\}, \]
\[ \hat y_0'=\{i(V_0q_-^{1/2}(0)-q_+^{1/2}(0))y_0^{(n)},\ y_0^{[2n]},\ y_0^{[2n-1]},\ldots,\ y_0^{[n+1]}\}, \tag{15} \]
\(V_0\) is an arbitrarily fixed isometric operator mapping \(H_0^-\) onto \(H_0^+\). If, however, \(\dim H=\infty\) and \(n\ge1\), then separated self-adjoint conditions always exist and have the same form (12), but now one must, generally speaking, take
\[ \hat y_0=\{y_0^+ + V_1y_0^-,\ y_0,\ y_0',\ldots,\ y_0^{(n-2)}\}\in H^{n+}, \]
\[ \hat y_0'=\{iV_1y_0^- - iy_0^+,\ y_0^{[2n]},\ y_0^{[2n-1]},\ldots,\ y_0^{[n+2]}\}\in H^{n+}, \tag{16} \]
where
\[ y_0^+=\{q_+^{1/2}(0)y_0^{(n)},\ \tfrac12(y_0^{[n+1]}-iy_0^{(n-1)})\}\in H_0^+\oplus H, \]
\[ y_0^-=\{q_-^{1/2}(0)y_0^{(n)},\ \tfrac12(y_0^{[n+1]}+iy_0^{(n-1)})\}\in H_0^-\oplus H, \]
\(V_1\) is an arbitrarily fixed isometric operator mapping \(H_0^-\oplus H\) onto \(H_0^+\oplus H\). If, in particular, \(\dim H_0^+=\dim H_0^-\), then here too one may use formulas (15) instead of (16).
Corollary. For odd \(\dim H<\infty\), for the operation \(l[y]\) (13) of odd order there do not exist separated self-adjoint boundary conditions.
Example 3. For the operation \(i(P-P^\perp)\dfrac d{dt}y\), where \(P\) and \(P^\perp\) are orthoprojectors onto \(H_1\subseteq H\) and onto \(H_1^\perp\), separated self-adjoint boundary conditions exist if \(\dim H_1=\dim H_1^\perp\), and in this case reduce to the form \(Py(0)=V_1P^\perp y(0)\), \(Py(b)=V_2P^\perp y(b)\), where \(V_1,V_2\) are arbitrary isometric operators mapping \(H_1^\perp\) onto \(H_1\).
Example 4. Generalized periodic conditions for the operation \(l[y]\) (13) of order \(2n+1\):
\[ y_b^{[2n-k]}=U_ky_0^{[2n-k]},\qquad y_b^{(k)}=U_ky_0^{(k)},\qquad q_\pm^{1/2}(b)y_b^{(n)} \]
\[ =V^\pm q_\pm^{1/2}(0)y_0^{(n)},\qquad k=0,1,\ldots,n-1, \]
\(U_k\) are unitary operators in \(H\), and \(V^\pm\) isometrically map \(H_0^\pm\) onto \(H_b^\pm\), respectively.
Physico-Technical Institute of Low Temperatures
Academy of Sciences of the Ukrainian SSR
Received
6 VI 1968
CITED LITERATURE
- M. G. Krein, Mat. sbornik, 21, 3, 365 (1947).
- N. I. Akhiezer, I. M. Glazman, Theory of Linear Operators, Moscow, 1966.
- M. A. Naimark, Linear Differential Operators, Moscow, 1954.
- E. A. Coddington, N. Levinson, Theory of Ordinary Differential Equations, IL, 1958.
- M. I. Vishik, Tr. Mosk. matem. obshch., 1, 187 (1952).
- M. G. Krein, DAN, 74, No. 1, 9 (1950).
- F. S. Rofe-Beketov, DAN, 156, No. 5, 1029 (1964).
- M. L. Gorbachuk, Ukr. matem. zhurn., 18, No. 2, 3 (1966).