Convolution equations and singular integral operators. Selected papers of Israel Gohberg and Georg Heinig, Israel Gohberg and Nahum Krupnik. Transl. from the Russian by Oleksiy Karlovych. Edited by Leonid Lerer, Vadim Olshevsky and Ilya M. Spitkovsky (Q1036128): Difference between revisions

This volume contains English translations, prepared by A.~Karlovich, of 5~papers by I.~Gohberg and G.~Heinig and of 8~papers by I.~Gohberg and N.~Krupnik which appeared in Russian in the years from 1968 to 1976. It also contains a very competent, instructive, and beautifully written commentary on these papers by L.~Lerer, V.~Olshevsky, and I.~Spitkovsky. For about 100 years, four classes of operators, namely, Toeplitz operators, Wiener-Hopf operators, singular integral operators, and the operators arising from Riemann-Hilbert problems, have played a prominent role in concrete operator theory. The reason for the enduring interest in these operators comes from countless really important applications on the one hand and the fact that these operators are the best understood non-selfadjoint operators on the other. The four classes of operators are closely related, one could say they are nothing but one and the same thing in different guises, although this was fully realized only in the 1960s, in part also due to the work of the authors of this volume. It took half a century and the efforts of many great mathematicians, including, for instance, F.~Noether, J.~Plemelj, S.~G.~Mikhlin, G.~Fichera, T.~Carleman, to understand these operators in the case of continuous symbols or coefficients, and the very last chapter of this story was written by I.~Gohberg and M.~G.~Krein in the 1950s. The challenges for further research were discontinuous symbols and coefficients as well as questions concerning concrete work with these operators, for example, their effective inversion. Questions of the last kind were the starting point of the work by Gohberg and Heinig. They considered the question of inverting an \(n \times n\) Toeplitz matrix \(T_n=(a_{j-k})_{j,k=1}^n\). It is a common opinion that the actual problem is the solution of equations and not the construction of the inverse matrix. That's right in many instances, but things are different in the case of Toeplitz matrices. Previous work by \textit{W. E. Trench} [J. Soc. Ind. Appl. Math. 12, 515--522 (1964; Zbl 0131.36002)] and \textit{I. C. Gohberg} and \textit{A. A. Semencul} [Mat. Issled. 7, No.~2(24), 201--223 (1972; Zbl 0288.15004)] contained formulas for \(T_n^{-1}\) which allowed the superfast computation of \(x=T_n^{-1}y\). These formulas were based on several additional assumptions. In 1970, Heinig went to Kishinev (now Chisinau) to write his dissertation under the supervision of Gohberg. The topic given to him by Gohberg was first to remove the existing extra assumptions and secondly, to extend the formulas to block Toeplitz matrices, that is, to the case where the entries \(a_{j-k}\) are themselves matrices. All these problems were then solved and resulted in the first two papers contained in the volume under review. There are several ways to eliminate the extra assumptions, and accordingly about a dozen of papers by many authors were written on this issue in the years after the Gohberg-Heinig papers. More importantly, the work by Gohberg and Heinig already contained the germs of what later became the concept of the displacement rank and thus of an extremely useful tool for studying structured matrices, and this work also laid the foundations for the many years of joint work of Heinig and K.~Rost in structured matrices. Finally, in the 1990s, the formulas were thoroughly checked for numerical accuracy and stability, and it turned out that they are not only fast but also accurate and stable provided the condition number of \(T_n\) is not too large. The third Gohberg-Heinig paper in this volume [Rev.\ Roum.\ Math.\ Pures Appl.\ 20, 55--73 (1975; Zbl 0327.45009)] deals with the continuous analogue of Toeplitz matrices, that is, with the inversion of integral operators with matrix-valued kernels of the form \(k(x-y)\) over finite intervals. Having the inverse operator, one has good chances to solve certain inverse problems. Consequently, Gohberg and Heinig embarked on inverse problems in their papers. In the course of these investigations, they encountered resultant matrices. Given two polynomials of degree \(m\) and \(n\), their resultant matrix is a matrix \(R\) of order \(m+n\) such that \(m+n-\mathrm{rank}\, R\) is the number of common zeros of the polynomials. The last two papers by Gohberg and Heinig [Acta Sci.\ Math.\ 37, 41--61 (1975; Zbl 0298.15013); Acta Math.\ Acad.\ Sci.\ Hung.\ 28, 189--209 (1976; Zbl 0341.15011)] in this volume are devoted to constructing analogues of the resultant matrix for matrix polynomials and for matrix-valued entire functions. To solve the latter problem, they replace the resultant matrix by a cleverly constructed resolvent integral operator. Their results originated a big development in the spectral theory of matrix polynomials that is lasting to date and whose main actors include A.~Markus, P.~Lancaster, L.~Rodman, M.~A.~Kaashoek, and L.~Lerer. Let us turn to the eight papers by Gohberg and Krupnik, which arose from their fruitful collaboration in Kishinev in the late 1960s and early 1970s. For the sake of definiteness, we state everything in the language of Toeplitz operators. The Toeplitz operator with the symbol \(a\) will be denoted by \(T(a)\). The symbol is a function in \(L^\infty\) on a curve \(\Gamma\) and \(T(a)\) acts on a weighted Hardy space \(H^p(\Gamma,w)\) by the rule \(f \mapsto P(af)\), where \(P\) is the Riesz projection of \(L^p(\Gamma,w)\) onto \(H^p(\Gamma,w)\). We have \(P=(I+S)/2\) with the Cauchy singular integral operator \(S\) over \(\Gamma\). If \(L^p(\Gamma,w)=L^2({\mathbb{T}})\), then the matrix representation of \(T(a)\) in the standard orthonormal basis of \(H^2({\mathbb{T}})\) is just the infinite Toeplitz matrix \((a_{j-k})_{j,k=1}^\infty\). To understand the early Gohberg-Krupnik papers, it is essential to have a feel for the spirit in the field in the early 1960s. As already said, everything was clear for continuous symbols. Nevertheless, in 1959, \textit{A.~Calderón, F.~Spitzer}, and \textit{H.~Widom} [Ill.\ J.\ Math.\ 3, 490--498 (1959; Zbl 0091.11101)] established an invertibility criterion for Toeplitz operators with symbols from the Wiener algebra, and only later they learned that a substantial part of their results are already in a 1958 paper by M.~G.~Krein. A new development got under way around 1960 by H.~Widom and I.~B.~Simonenko who (independently) proved criteria for the invertiblity and Fredholmness of \(T(a)\) for general \(a \in L^\infty\) in terms of certain factorizations of \(a\). These criteria solved the problem in analytical language, but what one was actually looking for were criteria in geometrical language. A first new result in this vein was also established by Widom and Simonenko (again independently), namely that the spectrum of \(T(a)\) on \(H^2({\mathbb{T}})\) is contained in the convex hull of the essential range of \(a\). This became later known as the Brown-Halmos theorem. To give another example, as early as 1961, Simonenko knew that the operator \(T(a)\) is invertible if and only if it is Fredholm of index zero. This result was rediscovered and first published by \textit{L.~Coburn} [Mich.\ Math.\ J.\ 13, 285--288 (1966; Zbl 0173.42904)] and then went into the textbooks as Coburn's lemma. Thus, those times were characterized by the usual chaotic behavior of hot fields of research after the big bang. Two groundbreaking papers were published by Widom in 1960. In the first, he considered Toeplitz operators \(T(a)\) on \(H^2({\mathbb{T}})\) with piecewise continuous symbols \(a\) having at most finitely many jumps and showed that the essential spectrum of \(T(a)\) is the continuous curve which results from the essential range of \(a\) by filling in the line segments \([a(t-0),a(t+0)]\) between the endpoints of each jump. In the second paper, Widom carried the results over to Toeplitz operators on \(H^p({\mathbb{R}})\) and proved that in this case one has to fill in certain circular arcs \({\mathcal A}_p(a(x-0),a(x+0))\) for jumps at \(x \in {\mathbb{R}}\) and \({\mathcal A}_q(a(+\infty),a(-\infty))\) for the possible jump at infinity. The subscripts \(p\) and \(q\) indicate that the shape of the arcs depends on \(p\), and \(q\) is nothing but \(p/(p-1)\). Widom gives credit to B.~V.~Khvedelidze, who had similar results in special cases but in non-geometrical language earlier. However, the circular arcs were born with Widom's 1960 paper. The first Gohberg-Krupnik paper in the volume under review [Stud.\ Math.\ 31, 347--362 (1968; Zbl 0179.19701)] was published in 1968 and contains the analogue of Widom's results for Toeplitz operators on \(H^p({\mathbb{T}})\). Gohberg and Krupnik cite Widom's second paper and their approach is also based on results by Khvedelidze. True progress is in their second paper [Mat.\ Issled.\ 4, No.~3, 54--62 (1969; Zbl 0254.47045)]. There they show that the Banach algebra generated by Toeplitz operators on \(H^p({\mathbb{T}})\) with piecewise continuous symbols is commutative modulo the compact operators, they identify the maximal ideal space as the cylinder \({\mathbb{T}} \times [0,1]\) with an exotic topology, they determine the Gelfand transform, and they also establish an index formula for the Fredholm operators in the Banach algebra. In their third paper [Mat.\ Issled.\ 5, No.~3(17), 46--57 (1970; Zbl 0228.47037)], Gohberg and Krupnik observe that certain integral operators with logarithmic singularities in the kernel can be treated by having recourse to their results on the Banach algebra generated by the Toeplitz operators with piecewise continuous symbols. The fourth paper by Gohberg and Krupnik [Mat.\ Issled.\ 5, No.~2(16), 89--103 (1970; Zbl 0223.45005)] is devoted to the case where \(\Gamma\) is a so-called composed curve, that is, a finite union of simple curves. In this setting, things change dramatically because \(S^2\) is no longer the identity operator \(I\) and still worse, \(S^2-I\) is not even compact. Nowadays the case of composed curves is tackled by using \(N\)-projections theorems, but as those theorems were not yet available, Gohberg and Krupnik had to invent something new to master the difficulty. They did it by replacing scalar symbols with matrix symbols. In this paper, they also admit power weights \(w\). The effect is that the Gelfand map on the cylinder \({\mathbb{T}} \times [0,1]\) changes on a finite number of segments \(\{t_j\} \times [0,1]\). The fifth Gohberg-Krupnik paper [An. Ştiint. Univ. Al. I. Cuza Iaşi, Ser. Nouă, Secţ. I\,a 19, 43--71 (1973; Zbl 0437.47019)], published in 1973, is another milestone. It contains what is now called the local principle of Gohberg and Krupnik. That the Fredholmness of Toeplitz operators is local in nature was already understood by Simonenko in the very early 1960s. For example, he then showed that if \(a \in L^\infty\) and if for every \(t \in \Gamma\) there exist an open arc \(U_t \subset \Gamma\) containing \(t\) and a function \(a_t \in L^\infty\) such that \(a|U_t =a_t|U_t\) and \(T(a_t)\) is Fredholm, then \(T(a)\) itself is also Fredholm. In 1965, Simonenko generalized this beautiful observation to his ``local principle''. Gohberg and Krupnik reformulated the local principle and stated it in a form that made it become one of the most popular tools for studying invertibility in (non-commutative) Banach algebras. There are close connections between Gohberg and Krupnik's local principle and other localization techniques, such as localization over central subalgebras, but we will not embark on this issue here. Using their local principle, Gohberg and Krupnik were able to give extremely lucid proofs of many previous results. In the paper at hand, they demonstrate this in particular by extending results of R.~V.~Duduchava on Toeplitz operators with piecewise continuous symbols on \(\ell^p\) to the case of symbols with countably many jumps and to the Banach algebra generated by such operators. In the last three papers of the volume, Gohberg and Krupnik deepen their results for composed curves and consider singular integral operators with involutive shifts. The highlight of the last two papers is the purely algebraic clarification of the fact that the Fredholmness of an operator with a shift is equivalent to the Fredholmness of a certain doubled operator without shift. The doubling does not cause problems, because it merely transforms matrix-valued coefficients of the order \(n\) to coefficients of the order \(2n\), and matrix coefficients had never been something that Gohberg and Krupnik didn't like. The impact of the work by Gohberg and Krupnik on subsequent research is enormous. We confine ourselves to three topics. First, the question whether in a sequence \(\{A_n\}_{n=1}^\infty\) of matrices all matrices \(A_n\) are invertible for \(n \geq n_0\) and satisfy \(\|A_n^{-1}\| \leq M < \infty\) is of fundamental importance in numerical analysis. In the case where \(A_n\) is the \(n \times n\) truncation of an infinite Toeplitz matrix with a discontinuous symbol, this question had been studied by many authors, including I.~Gohberg and I.~A.~Feldman, I.~E.~Verbitsky, and N.~Krupnik. However, a satisfactory solution was not in sight. In 1980, B.~Silbermann turned to the problem, and inspired by the local principle of Gohberg and Krupnik, he became to understand how to reformulate the question in terms of invertibility in an ingeniously constructed Banach algebra. In this way, he disposed of several problems that were open at that time, and since then localization techniques have entered numerical analysis and proven to be of immense use in proving the stability of approximation methods and numerical algorithms. Secondly, the consideration of composed curves led to a remarkable development in \(N\)-projections theorems. In the 1970s, R.~G.~Douglas discovered a very elegant way of constructing a symbol calculus for Toeplitz operators on \(H^2({\mathbb{T}})\) with the help of Halmos's two projections theorem. This is a theorem for Hilbert space operators. To proceed in this way for Toeplitz operators on \(H^p({\mathbb{T}})\), one needed such a theorem for two projections on Banach spaces or more generally, for two idempotents in Banach algebras. In the late 1980s and early 1990s, S.~Roch and B.~Silbermann and independently Gohberg and Krupnik were indeed able to establish such a theorem. Two projections are enough on curves like the unit circle \({\mathbb{T}}\), because each point has the valency two, an incoming and an outgoing arc. However, points of composed curves may have arbitrary valencies, and this called for an appropriate \(N\)-projections theorem. We remark that we are not speaking of \(N\) arbitrary projections. The nature of singular integral operators on composed curves implies that the projections are subject to additional restrictions or axioms. The theorem that covered exactly the cases of interest was found in 1996 by the joint efforts of Roch, Silbermann, Gohberg, Krupnik, I.~Spitkovsky, Yu.~I.~Karlovich, and the reviewer. Since then the theory of operators on composed curves is a pleasure when compared with the forced march Gohberg and Krupnik had to go in the 1970s. Thirdly, Widom, Gohberg, and Krupnik considered the operators on Lyapunov curves \(\Gamma\) with power weights \(w\), which led to the appearance of circular arcs in the spectra. In the 1990s, I.~Spitkovsky, Yu.~I.~Karlovich, and the reviewer passed to general Ahlfors-David-Carleson curves \(\Gamma\) and general Muckenhoupt weights \(w\). This is technically very exacting, and had it turned out that the final results remain essentially the same, this would have been a thankless job. Fortunately, it was discovered that more general curves and weights yield a fantastic metamorphosis: the circular arcs first become logarithmic double-spirals, then blow up to horns and logarithmic horns, and eventually we see logarithmic leaves with a halo and are at the end of an exciting story.