Hilbert Spaces Fall 2001 Final Exam

  1. (This is based on B, p. 171, 12(xii).)
    a) Completely describe the spectrum of the left shift operator on l2. Which spectral values, if any, are eigenvalues?

    b) Show that the spectrum of the right shift operator on l2 is the entire disc {|z| <= 1}. Which spectral values, if any, are eigenvalues?

    Solution

    a) Let L denote the left shift operator, so Since L is a contraction, the spectrum is a subset of {z : |z| <= 1} (see class notes). I claim that if |z| < 1, then z is an eigenvalue of L. Indeed, from the eigenvalue equation (L s = z s) and the formula given above, It follows that the eigenvectors corresponding to z are multiples of This is in l2 iff |z| < 1.

    Every z with |z| < 1 is thus an eigenvalue. Because the spectrum is a closed set, sp(L) = {z : |z| <= 1}.

    b) It is shown in B that the right shift R = L* has no eigenvalues. However, its spectrum is the same as that of L for the following reason. If |z| < 1, then the closure of Ran(R - z) is the orthogonal complement of N(R* - z*) = N(L - z*) = the multiples of the eigenvector of L with eigenvalue z*.

    Since dim(N(L - z*)) = 1, Ran(R - z) is not all of l2, so (R - z)-1 is not in B(l2). This establishes that sp(R) contains {z : |z| < 1}. Since R is an isometry, sp(R) lies inside {z : |z| <= 1}, and since the spectrum is closed it is precisely {z : |z| <= 1}.

    Note: If you suspect that there is a general rule for bounded operators that sp(A*) = (sp(A))*, you are correct (and this is part (ii) of the same problem in B). This is not the situation for eigenvalues, as this problem teaches us..

  2. Assume that A is positive and bounded. Show that sqrt(A) is compact if and only if all nonzero AP values of A are isolated eigenvalues of finite multiplicity.

    Solution

    The Riesz-Schauder theorem tells us that the nonzero part of the spectrum of a compact operator consists of isolated1 eigenvalues of finite multiplicity, so for the "only if" part it suffices to show that A is compact. Since A = sqrt(A) sqrt(A)*, this follows form the first TFAE of the class notes on compact operators.

    The other direction is the tricky one. Suppose that the eigenvalues of A have the stated property. I shall show that A is compact, and it will follow that sqrt(A) is compact because of the first TFAE of the class notes on compact operators.

    Since A = A*, the spectrum consists entirely of AP values. By hypothesis these are eigenvalues zn of finite multiplicity without accumulation (adherent) points other than possibly 0. If en are the orthonormalized eigenvectors, we would like to show that A = Sum(zn Pn), where Pn f := <f | en> en; since there are either a finite number of zn or else they have a limit of 0, this operator is compact.

    Let B := A - Sum(zn Pn). B has no nonzero eigenvalues, because if it did, then A would have an eigenvalue we have not accounted for. Moreover, B en = 0. If z is a nonzero AP value for B, then we may assume by projecting away all the en that the vectors fk, ||fk|| = 1, such that

    are orthogonal to all en. Hence Pn fk = 0, and it follows that Consequently z is also a nonzero AP value of A, contradicting our hypothesis that that we have accounted for all such values already with the zn. Conclusion: The only AP value of B is 0. By B, p. 170, Thm. 2, ||B||op = 0, so A = Sum(zn Pn) as claimed, which is compact.
    1. Being "isolated" is the same as not being an accumulation (= adherent) point.
  3. Consider the following sets of numbers and kinds of operators on a classical Hilbert space:
    1. {1/n}, n = 1, 2, ...
    2. The unit circle in the complex plane
    3. {1, 0}
    4. {x: 0 <= x <= 2}
    5. {n/(n2 + i}, n = 0, 1, ...

    1. self-adjoint operators
    2. positive operators
    3. orthogonal projectors (what B calls "projections")
    4. unitary operators
    5. compact operators (what B calls "completely continuous" operators)
    6. contractions

    In the following grid write S if the set is possibly the spectrum of the kind of operator, and write E it the set is possibly the entire set of the eigenvalues of the operator:

    Solution

    IIIIIIIVV
    a)EXE,SSX
    b)EXE,SSX
    c)XXE,SXX
    d)XSXXX
    e)EXE,SXE,S
    f)ESE,SXE,S