Abstract
Analysis of the asymptotic behaviour of the spectral characteristics of Toeplitz matrices as the dimension of the matrix tends to infinity has a history of over 100 years. For instance, quite a number of versions of Szegő's theorem on the asymptotic behaviour of eigenvalues and of the so-called strong Szegő theorem on the asymptotic behaviour of the determinants of Toeplitz matrices are known. Starting in the 1950s, the asymptotics of the maximum and minimum eigenvalues were actively investigated. However, investigation of the individual asymptotics of all the eigenvalues and eigenvectors of Toeplitz matrices started only quite recently: the first papers on this subject were published in 2009–2010. A survey of this new field is presented here.
Bibliography: 55 titles.
Export citation and abstract BibTeX RIS
.
§ 1. Introduction
Let be a Lebesgue integrable function on the unit circle and let denote the Toeplitz matrix formed by the Fourier coefficients of :
Then is called the symbol of the sequence of matrices .
The behaviour of various spectral characteristics of Toeplitz matrices (eigenvalues and singular values, eigenvectors, determinants, condition numbers and so on) has been an object of active investigation for a century, starting with Szegő's paper [1] (see also the books [2]–[6] and the references there). First and foremost note the numerous versions of Szegő's theorem on the asymptotic distribution of the eigenvalues and the Avram-Parter-type theorems on the asymptotic behaviour of the singular values (see [7]–[12]). There is extensive literature on the asymptotic behaviour of the determinants of Toeplitz matrices (see the monographs [4]–[6] and the papers [13]–[19] and the references there). The asymptotics of the maximum and minimum eigenvalues has also been a focus of research (see [20]–[25]). The shape of the limit set of the eigenvalues and its stability have been investigated (see [26]–[28] and [15]). Note that the latter papers (by contrast with most of the papers cited before them) concern the more complicated case of essentially complex-valued symbols, which has not yet been analyzed in full generality.
This considerable interest in the asymptotic behaviour of the spectral characteristics of large Toeplitz matrices was motivated to a significant extent by many important applications: stochastic processes and time series analysis (see [2]); signal processing [29]; the numerical solution of differential and integral equations [30]; image processing [31]; quantum mechanics [32]. We point out the survey paper [19], which discusses the influence of the Ising model in statistical mechanics on the development of the asymptotic theory of the determinants of Toeplitz matrices.
Despite the considerable interest in this field that many authors have displayed, the individual asymptotic behaviour of all eigenvalues and eigenvectors was hardly investigated before 2008. In this connection we can only mention the well-known case of tridiagonal Toeplitz matrices (for instance, see [2] and [5]) and the case of linearly growing matrix elements considered recently in [33], where the eigenvalues and eigenvectors are calculated explicitly. On the other hand the importance of this problem is beyond doubt. For example, many papers are devoted to numerical methods for finding the spectrum of Toeplitz matrices of large size (see [34], [35] and the references there). For dimensions of order – this problem can be solved effectively using modern hardware and the algorithms described in the above papers. However, in statistical physics and some other applied problems we encounter dimensions of order –, when no alternative to the asymptotic method exists. In addition, asymptotic formulae give more precise information on the local structure of the set of eigenvalues, the distances between them, accumulation points, their dependence on physical parameters and so on.
In our opinion, the main difficulties arising in the implementation of the asymptotic method are as follows. First, the distances between successive eigenvalues are small (very small when we are in a neighbourhood of an accumulation point!) and our asymptotic formulae must be able to `separate' them. Furthermore, in addition to (the dimension of the matrix) being large, the problem involves another parameter, the index of the eigenvalue. So the asymptotic expansions with respect to , as goes to infinity, which we construct must hold uniformly in , . Finally, we stress that the determinants that appear in the eigenvalue problem in a natural way are related to one of the most complicated cases, when the symbol of the Toeplitz matrix has zeros on the unit circle.
In the papers [36]–[38], published in 2009–2010, asymptotic representations were constructed in the case of real-valued polynomial symbols (in other words, for Hermitian Toeplitz matrices with a finite number of nontrivial diagonals) that satisfy the so-called SL (simple loop) condition. This means that the real-valued symbol has precisely one minimum and one maximum on the unit circle (the precise definition is expressed by conditions (i) and (ii) in §2). The condition ensures a certain regularity in the arrangement of the eigenvalues; for instance, there are no multiple eigenvalues. In [36] and [37], based on Widom's well-known formula for the determinant of a Toeplitz matrix with polynomial symbol (see [21] and also [5], §2.4) we found an equation for the eigenvalues, with an exponentially small remainder term, and performed an asymptotic analysis of this equation. In [38] we also used explicit formulae from [39] to calculate eigenvectors of matrices with polynomial symbols, and on this basis we wrote down asymptotic formulae with a remainder term which is exponentially small in .
In the paper [40] published in 2012 the authors considered the case of infinitely smooth symbols also satisfying the SL condition. Using the asymptotic formula for determinants with Fisher-Hartwig singularities obtained in [18] the authors obtained the same nonlinear equation as in [36] and [37], but with remainder of the form . Next they deduced two-sided estimates for the distance between arbitrary pairs of successive eigenvalues. Before the statement of this theorem the authors mentioned that the smoothness conditions imposed on the symbol can be relaxed.
In [41] and [42] we considered an -symbol whose fourth derivative exists and has a certain regularity. We used a method distinct from [37] and [40]. It is based on a precise equation for the eigenvalues, which we derived, and on an asymptotic analysis of it. This equation is given in terms of the inverse matrix of a certain Toeplitz matrix of dimension with positive symbol equal to the ratio of the difference (where is the spectral parameter) and a second-order polynomial with the same zeros as this difference. We also point out [43], where the symbol is only assumed to have a first derivative and to satisfy certain additional conditions near the minimum and maximum points.
In [44] and [45] we abandoned the selfadjoint case and investigated Toeplitz matrices with complex-valued symbols. These symbols have a special feature: the corresponding function takes the unit circle to a curve without interior. In other words, the unit circle covers its image with multiplicity two by going in the `direct' and `reverse' direction along it. It is well known that then the limit set of the sequence of spectra coincides with this image curve, so for large the eigenvalues lie close to this curve. Again, as in [36] and [37], in these papers we used known explicit formulae for determinants and inverse matrices because our symbols were polynomials. The resulting equation for the eigenvalues has the same structure as in the real-valued case; however, it must be considered in the complex domain, rather than on an interval of the real line, which brings in quite a number of significant technical difficulties.
In 2009–2010 Kadanoff and his collaborators published the papers [46] and [47], where they considered a symbol, with power-like singularities, in the Fisher-Hartwig class. They presented formulae for the eigenvalues and eigenvectors. These formulae were deduced using heuristic arguments and justified by numerical experiment. In the same period of time, in 2008–2010 we considered a certain subclass of Fisher-Hartwig symbols. In this way, in [48]–[50] we refined and justified the formulae in [46] and [47] rigorously for symbols in the intersection of the classes under consideration.
We also point out the paper [51] close to this field. It looks at the asymptotic behaviour of the eigenvalues and eigenvectors of the Wiener-Hopf integral operator on a finite interval as the length of the interval tends to infinity, and it deals with the case of a symmetric rational (generally speaking, complex-valued) symbol. This problem is related to the same problem for Toeplitz matrices; however, its peculiarity consists in there being an infinite number of eigenvalues and eigenfunctions of the operator for an interval of any length.
Finally, we mention [52] and [53], where we go over from convergence in the sense of distributions (Szegő's theorem) to uniform convergence of the quantile function. For rather wide classes of symbols, approximation of eigenvalues can be expressed in terms of this function.
In §2 we state results for -symbols based on [43] and [42]. Toeplitz matrices with symmetric symbols are the subject of §3, where we present the main results of [44] and [45]. In §4 we look at Fisher-Hartwig symbols and discuss results from [48] and [50]. In §5 we investigate Wiener-Hopf operators on a finite interval (see [51]).
§ 2. SL-symbols
2.1. Eigenvalues
Assume that and let denote the weighted Wiener algebra of functions with representations
where the Fourier coefficients satisfy
Along with we look at the function given by .
We define the class as the set of symbols in such that
- (i)is real-valued;
- (ii)the range of is an interval where ; , , and there exists a point such that , for , for and .
Note that (i) is equivalent to all the matrices , , being Hermitian (selfadjoint). If , then , where is the integer part of . Thus, if with then, in particular, belongs to . Furthermore, in (ii) we assume that has second derivatives at the points , and .
We will say that the symbol belongs to the class with if and the following additional condition is satisfied:
- (iii)there exist functions such that
By Lemma 3.1 in [41], if , then the functions and defined by (2.1) must belong to . In the definition of the class we impose the stronger condition (iii), that is, we assume that the symbol has additional smoothness at the points where it takes the minimum and maximum values. Using the same lemma we can show that each has the representations
with .
Let , . For each there exists precisely one point on the interval such that and precisely one point such that . In other words and are the inverse functions of restricted to and , respectively. For each , takes values not exceeding on the intervals and . Let denote the arithmetic mean of the lengths of these two intervals:
where denotes Lebesgue measure on . Note that is continuous and bijective and let denote its inverse function. Up to a linear change of the argument, and are the distribution function and the quantile function of the `random variable' on the probability space with normalized Lebesgue measure. Set
Note that the derivatives of and are unbounded in a neighbourhood of and , whereas the functions and have continuous first derivatives on the whole of . In addition, it is easy to see that
For each the function has two zeros, and . We can define a positive function in terms of these:
Let be defined by
The singular integrals above are taken in the sense of the Cauchy principal value.
Let denote the eigenvalues of the matrix . We can now state the main results in [43].
Theorem 2.1. Let and . Then for each :
(i) all eigenvalues of are distinct, so that ;
(ii) the quantities , , satisfy the equations
where uniformly in as ;
(iii) for sufficiently large , equation (2.5) has a unique solution for each .
Set .
Theorem 2.2. If all the assumptions of Theorem 2.1 are fulfilled, then
where uniformly in as . The coefficients can be calculated explicitly; in particular,
Theorem 2.3. If all the assumptions of Theorem 2.1 are fulfilled, then
where for and for uniformly in as . The coefficients can be calculated explicitly; in particular,
Remark 2.1. Let and . Then the eigenvalues (for which ) have the following asymptotics:
This means that the distance between two successive eigenvalues has the two-sided estimate
where the constants and are independent of and . On the other hand the difference also has order . In this connection an approximation
is often not sufficient for computations because the approximate formula (2.7) does not `separate' eigenvalues: a point need not lie closest to the th eigenvalue, but perhaps to some other eigenvalue, for instance, .
The assertion of Theorem 2.3 holds for all eigenvalues of . However, bearing in mind the behaviour of and close to the endpoints of (see Lemmas 4.2 and 4.6 in [43]) we can refine the asymptotic formulae for the extreme eigenvalues (cf. Corollary 2.5 in [41]).
(i) If , then
where , and as .
(ii) If , then
where , and as .
Theorem 2.1 and, as a consequence, also Theorems 2.2 and 2.3 are based on the following result, which provides an exact equation for the eigenvalues. We introduce some further notation. As functions of the form (2.3) are positive, the operator is invertible for each . Let be defined by
Here the Toeplitz matrix and the inverse matrix are realized as operators in the space
of polynomials of degree at most and the function is identically equal to one: . In other words, is a polynomial of degree at most constructed from the entries in the zeroth column of . For brevity we set , , so that we take to be treated as .
Theorem 2.5. Assume that , and let and . Then is an eigenvalue of the operator if and only if
Note that (2.10) involves the action of the inverse operator , so it is rather difficult to analyze directly. However, it can be analyzed asymptotically as , when we replace by the function , where is the inverse of the (infinite) Toeplitz operator. As is easy to calculate in terms of the Wiener-Hopf factorization of , we obtain equation (2.5) in this way.
We conclude this subsection by observing that the formulae in Theorem 2.3 have proved to be numerically highly efficient and yield nice results for (see [37], [41] and [43]).
2.2. Eigenvectors
The arguments resulting in the exact equation (2.10) enable us to deduce exact formulae for the components of the eigenvectors too. The results presented in this subsection were published in [42]. Keeping the notation used in §2.1, for brevity we write
where we recall that is related to the th eigenvalue of the matrix by .
Theorem 2.6. Assume that and let , and . Then the vector
whose th component, , is given by
is an eigenvector corresponding to the eigenvalue . In addition, is conjugation symmetric: .
That is, the eigenvector can be represented as a sum of three vectors: the leading term , the term , concentrated in a neighbourhood of the left-hand endpoint (for small its components exceed significantly the other terms in absolute value) and the term , concentrated in a neighbourhood of the right-hand endpoint. In other words, yields a rough approximation of the eigenvector, which is satisfactory for central components, while and play the roles of left- and right-hand corrections, respectively.
In the symmetric case, that is, when the symbol has the property (which is equivalent to ), the statement of Theorem 2.6 can be simplified slightly.
Corollary 2.1. Assume that and let be a function with real Fourier coefficients (so that ). Then for each
and the formulae in Theorem 2.6 can be written as follows:
Theorem 2.6 reveals the structure of eigenvectors, but can only be implemented if we calculate the exact eigenvalues and can invert -matrices. To simplify the formulae in Theorem 2.6 we look at the `basic' approximation (see Theorem 2.1) of the equation for the eigenvalues,
Note that if it has roots , then
uniformly in . Equation (2.13) is easy to solve numerically. In addition, asymptotic approximations to its solution are given by the formulae in Theorem 2.3. For we have the Wiener-Hopf factorization
where
Theorem 2.7. Let , . Let , , and for set
Let be the vector defined as the remainder term in the formula
Then as and this relation holds uniformly in and .
The integral involved in Theorem 2.7 can often be simplified. For example, if the symbol is rational, then the integral can be calculated using residue theory. In fact, in this case the function given by (2.3) can be expressed by
where and are polynomials in . It is easy to see that the symbol has a Wiener-Hopf factorization , where
here and are positive integers, and are formed from the zeros of , and are formed from the zeros of , and is a constant. Again, for brevity we write instead of .
Theorem 2.8. Let be a rational symbol in . Assume further that has only simple poles. Then an eigenvector can be expressed in the following form:
where, for ,
and for each the limit relation holds uniformly in and as .
Note that if some of the poles of are multiple, then we can also find an expression for using residue theory.
In [42] we presented numerical experiments validating the formulae in Theorems 2.7 and 2.8 and revealing the meaning of the different terms in these formulae.
§ 3. Symmetric symbols
In this section we consider complex-valued polynomial symbols of the form
where and . Such a polynomial is an even function of the angle :
Let denote the set of values of , that is, the curve on the complex plane formed by the points :
Then (for instance, see Theorems 5.28 and 5.32 in [3]) the limit as of the sequence of spectra of the coincides with . Thus, for sufficiently large the eigenvalues of the matrix lie in a small neighbourhood of .
3.1. Eigenvalues of symmetric Toeplitz matrices
Consider a symbol of the form (3.1) and assume that
- (1)the curve does not intersect itself; in particular, ;
- (2)for ;
- (3).
As usual, set . Then it follows from (3.2) that
For sufficiently small set , where
First we look at the equation for . It has roots in the complex plane, which we arrange in the order of ascending absolute values. Taking the symmetry (3.2) into account we can represent this set of roots as follows:
Thus, the root , , which lies outside the unit circle , corresponds to the root inside . By conditions 1 and 2 precisely two roots lie on :
The other roots are separated from uniformly in . Now let . It is easy to show that by 2 and 3, for sufficiently small the equation has a unique root in a neighbourhood of the upper half-circle, which has the form , where is an analytic function in . In accordance with this definition for . We will regard the as functions of and denote them by , where . Now there exists a positive such that
Consider the functions
and
It follows from the definition (3.6) of that for . Hence we can take a continuous branch of the function in the domain such that . We introduce the quantities
and the domains
where , ,
Note that is bounded uniformly in and .
Now we state the main result of this subsection.
Theorem 3.1. Let be a function of the form (3.1) such that conditions 1–3 are satisfied. Then the following results hold for sufficiently large .
(i) All the eigenvalues of are simple and for .
(ii) The points () such that satisfy
where and for some positive constant , uniformly in and as . Equation (3.11) has a unique solution in the domain .
(iii) Each equation of the form
has a unique solution in , and for some positive constant uniformly in as .
(iv) The function
is a contraction mapping of into itself; if
then
where and are defined by (3.10) and .
Setting in (3.15) and (3.16) we obtain the following asymptotic expansions for and .
Theorem 3.2. Let be a symbol of the form (3.1) such that conditions 1–3 are satisfied. Then
(i)
where as uniformly in ;
(ii)
where uniformly in as ,
We have given the asymptotic formulae (3.17) and (3.18) as examples. Using estimates (3.15) and (3.16) for we can find arbitrarily many terms of the expansions for and .
As a consequence of Theorem 3.2, we obtain a result on the extreme eigenvalues.
Theorem 3.3. Let be a symbol of the form (3.1) and assume that conditions 1–3 are satisfied. Then the following results hold:
(i) if , then
where , and as ;
(ii) if , then
where , and as .
Remark 3.1. The leading term of the asymptotic expansion (3.18) corresponds to a point on the curve . This approximation to the eigenvalue has accuracy and cannot be considered satisfactory: the distance between successive eigenvalues also has order . Thus, this approximation `does not separate' eigenvalues (see Remark 2.1). On the other hand the approximation has accuracy (see (3.16) for ) and can be treated as an individual approximation to the eigenvalue . In particular, the expression can be used to analyze the position of relative to the curve . In fact,
where
Setting
we obtain
That is, to within the eigenvalue lies on the normal to at the point . The distance from along the normal has order , provided that . Note that depending on the sign of the point lies to the right or the left of in the plane.
The numerical gain from the formulae in Theorems 3.1–3.3 was fairly comprehensively analyzed in [44].
3.2. Eigenvectors of symmetric Toeplitz matrices
Theorem 3.4. Let be a symbol of the form (3.1) and assume that conditions 1–3 are satisfied. Then eigenvectors associated with the eigenvalues can be chosen so that the following asymptotic expressions (putting for brevity) hold:
where the remainder estimate is uniform with respect to , and is independent of , and . Furthermore,
Remark 3.2. Theorem 3.4 was established as Theorem 2.3 in [45], albeit under the additional assumption that the function has only simple zeros for each in . Now, after [42], we can drop this technical condition.
Remark 3.3. Let us analyze the formulae in Theorem 3.4. As the root lies outside the unit circle, the term is small for components with large index , and the term is small when is large. In addition, it is easy to see that, as grows, the contribution of these terms can be significant for a few of the first and last components, respectively. Note however, that the number of these components does not depend significantly on . Components in the middle part of the vector are mostly well approximated by from (3.22). A numerical illustration of this observation can be found in [45].
Remark 3.4. In (3.22)–(3.24) we can replace the exact eigenvalues by their approximations (asymptotic expressions), for instance, the ones given by (3.18).
Remark 3.5. Formulae (3.22)–(3.24) become much simpler for eigenvectors corresponding to the extreme eigenvalues (see Theorem 2.4 in [45]).
§ 4. Eigenvalues and eigenvectors of Hessenberg Toeplitz matrices and a problem of Dai, Geary and Kadanoff's
4.1. Eigenvalues
Dai, Geary and Kadanoff [46] and Kadanoff [47] considered symbols of the form
where . They conjectured that for large the corresponding eigenvalues satisfy
where is some function (unknown to them). This conjecture was supported by numerical experiment and a certain heuristic argument.
Now we state the main results of [48] and compare these with (4.2).
As usual, let be the Hardy space of bounded analytic functions in the unit disc . For a fixed function let denote the number of windings of the image of , that is, the curve , about the point , and let be the set of points for which . In this subsection we treat matrices with symbols of the form , where
- (1.1)and ;
- (2.1), where and ;
- (3.1)extends analytically to an open neighbourhood of the set which does not contain ;
- (4.1)is a Jordan curve in and for each .
Note that if we set and in (4.1), then the symbol satisfies conditions
Let denote the determinant of . Then the eigenvalues of the matrix are the roots of the equation . By our assumptions is a Hessenberg matrix, that is, it can be obtained from a lower triangular matrix by adding one nontrivial diagonal. In combination with the Baxter-Schmidt formula (see [5], §2.3, for instance) this enables us to represent as a Fourier integral. Its value depends significantly on the pole at and the singularity of the term at . Let be a small neighbourhood of zero in . We can show that for each there exists a unique point such that . An asymptotic analysis of the Fourier integral just mentioned yields an asymptotic representation for .
In all the statements in this subsection we take
Theorem 4.1. Let be a symbol satisfying conditions
where as uniformly in .
We now turn to the main results in this subsection. Set . For the values of under consideration there exist positive integers and such that , but . Recall that .
Theorem 4.2. Let be a symbol with properties
where as uniformly in .
Thus formula (4.4) proves the conjecture (4.2) in the special case when . In addition, we have found the function
and have refined the order of the error term.
Taking the value of at the point (4.4) we obtain the following asymptotic expression for .
Theorem 4.3. Let be a symbol with properties
where as uniformly for between and .
Remark 4.1. Here we have written out only a few of the first terms of the asymptotic expansion. Note however that using our method an arbitrary number of terms in this expansion can be obtained.
A numerical illustration of Theorem 4.3 can be found in [48].
4.2. Eigenvectors
To present the formulae for eigenvectors we define a function on by
and let denote its th Fourier coefficient.
is an eigenvector of corresponding to the eigenvalue .
The next result describes the asymptotic behaviour of for large and .
holds, where
We stress that we only consider points in . In this domain the functions and are bounded and separated away from zero. Hence the , , are also bounded and separated away from zero there. Note also that since , tends to zero as .
Remark 4.2. We compare the above results on the asymptotics of eigenvectors and the results in [46] and [47] concerning symbols of the form (4.1).
We have already mentioned that for the function (4.1) becomes `our' symbol of the form , with . In [46] the authors conjectured that an eigenvector has the form
as . Theorem 4.2 refines this conjecture and proves it while giving a rigorous estimate for the error term. Note that
This shows that (4.8) corresponds to the first term on the right-hand side of (4.7) in the case when the ratio is close to zero. Otherwise the representation (4.8) can give a large error: the corresponding numerical examples were presented in [50].
Remark 4.3. Our asymptotic expansion (4.7) holds for outside a small neighbourhood of zero and for large . For small , for instance, , , the values of can be calculated using the relation between the Fourier coefficients of and . Here we obtain a triangular system of equations, which can be solved by operations, and taking we need operations. We can calculate the remaining components using asymptotic formulae (4.7), which also have order of complexity from a numerical standpoint.
§ 5. Spectral theory of Wiener-Hopf operators with symmetric complex kernels and rational symbols on a large interval
A truncated Wiener-Hopf operator has the form
We assume here that is a function in , so the integral operator in (5.1) is a compact Hilbert-Schmidt operator on for every . Let denote the spectrum of . As is compact, all points in are eigenvalues.
Here we make the two basic assumptions. First, is a symmetric complex function, that is for . Second,
the so-called symbol of the operator, is rational. These two assumptions are equivalent to a representation of the form
where the are complex numbers with and the are polynomials with complex coefficients. Since for all if and only if for all , the symbol is an even rational function, and we can write
where , , and for all and . To stress the dependence of on the symbol , following [54] we denote by .
Putting we extend the symbol by setting . As before, let be the essential image of , that is, . By our assumptions is an analytic curve in the plane such that as varies from to the value of the symbol ranges from to on this curve, and when varies further from to , the value of the symbol traverses from to in the reverse direction.
It was shown in [54] that in our case the limit of coincides with . So we will look for eigenvalues in a neighbourhood of . We assume in addition that is a curve without self-intersections.
With each positive number we associate the half-strip
where is fixed and sufficiently large. Consider a finite covering of the half axis by open intervals . For each interval we consider either the closed subintervals of of the form
or the closed intervals of the form
lying in (depending on the position of 1 in ); see (3.6) and (3.7) in [51]. In this way we obtain families of rectangles
The proofs of the main results in this section are based on an explicit formula (see [55]) for the Fredholm determinants
Note that for each the equation has precisely complex roots
We can number them so that and for .
Fix an open neighbourhood of . Then if is sufficiently large. Set . For consider the two functions
and set
The main result of this section is the following.
Theorem 5.1. Let be the closure of in and assume that for in all the roots are distinct. Then there exists such that the following results hold for all .
(i) If is an eigenvalue of such that for some , then .
(ii) For each the set contains a unique eigenvalue of , which has algebraic multiplicity 1.
(iii) The function
is a contraction mapping of into itself; for
the relation
holds uniformly in : there exist constants independent of and such that
(iv) The following asymptotic expression holds for the points 2:
where
(v) The following asymptotic expression holds for :
where
Remark 5.1. We have presented formula (5.4) as an example of an explicit calculation of a few of the first terms in the asymptotic expansion. Using our method we can find an arbitrary number of these terms.
In conclusion we present a result concerning eigenfunctions.
Theorem 5.2. Assume that all the numbers are distinct. Let be an eigenvalue of the operator such that all the roots are distinct. Then each eigenfunction of corresponding to has the following form:
and satisfies the relation with for all ; it can also be expressed by
Note that the coefficients in the last formulae can be calculated using a system of linear equations with known coefficients.
The numerical aspects of the above formulae have been discussed in [51].
Footnotes
- 1
See the definition of the function below.
- 2
This appears as in (i).