Abstract
We prove a weak version of Hardy’s uncertainty principle using properties of the prolate spheroidal wave functions. We describe the eigenvalues of the sum of a time limiting operator and a band limiting operator acting on \(L^2(\mathbb {R})\). A weak version of Hardy’s uncertainty principle follows from the asymptotic behavior of the largest eigenvalue as the time limit and the band limit approach infinity. An asymptotic formula for this eigenvalue is obtained from its well-known counterpart for the prolate integral operator.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The classical Hardy’s uncertainty principle is formulated as follows.
Theorem 1.1
Let \(a,b,M > 0\), and let \(f\) be a measurable function on \(\mathbb {R}\) such that
and
for all \(x,\xi \in \mathbb {R}\). If \(ab > 1\), then \(f = 0\).
Several proofs of this classical theorem are known e.g., [3–5, 9]. Typically, they use methods of complex analysis, and rely on somewhat indirect arguments. Our objective is to give a new and direct proof with methods of real analysis, however only of the following weaker result.
Theorem 1.2
Let \(a,b,M > 0\), and let \(f\) be a measurable function on \(\mathbb {R}\) such that
and
for all \(x,\xi \in \mathbb {R}\). If \(ab \geqslant 4\), then \(f = 0\).
We prove this weak version of Hardy’s uncertainty principle using properties of the prolate spheroidal wave functions (PSWFs), which appear e.g., in a solution of the concentration problem for bandlimited functions [7]. First, we describe the spectrum of the sum of a time limiting operator and a band limiting operator acting on \(L^2(\mathbb {R})\). Specifically, we express the spectrum in terms of the eigenvalues of the prolate integral operator.
Then we derive the weak version of Hardy’s uncertainty principle from the asymptotic behavior of the largest eigenvalue of the sum of the time and band limiting operators as the time limit and the band limit approach infinity. An asymptotic formula for this eigenvalue is obtained from its well-known counterpart for the prolate integral operator.
Our approach reveals a relationship between Hardy’s uncertainty principle and the theory of bandlimited functions.
2 Mathematical Preliminaries
The Fourier transform of a function \(f \in {\varvec{L}}^2 (\mathbb {R})\) is a bounded operator on \({\varvec{L}}^2 (\mathbb {R})\) defined as follows:
where the limit is taken in \({\varvec{L}}^2 (\mathbb {R})\). We also use the notation \(\hat{f}\) for \(\mathcal {F} f\).
\(\mathcal {F}\) is invertible on \({\varvec{L}}^2 (\mathbb {R})\), and its bounded inverse, the inverse Fourier transform, is defined as follows:
Consequently,
i.e., the Fourier transform is a unitary operator on \({\varvec{L}}^2 (\mathbb {R})\).
A bounded operator \(P\) on a Hilbert space is called idempotent if
A bounded operator \(P\) on a Hilbert space is called an orthogonal projection if it is idempotent and Hermitian, i.e.,
We denote the characteristic function of a set \(E \subset \mathbb {R}\) by \(\chi _E\), i.e.,
For a fixed set \(E \subset \mathbb {R}\), the mapping \(f \mapsto \chi _E f\) is an orthogonal projection on \({\varvec{L}}^2 (\mathbb {R})\). We also denote this projection by \(\chi _E\). In particular, we use the notation \(\chi _{(-\tau ,\tau )}\), when \(E = (-\tau ,\tau )\), \(\tau > 0\).
For a fixed \(\omega > 0\), we define the operator \(S_\omega \) as follows:
The operator \(S_\omega \) is also an orthogonal projection on \({\varvec{L}}^2 (\mathbb {R})\).
The integral kernel of \(S_\omega \) is
This kernel is computed explicitly as follows:
If \(P\) is an orthogonal projection, then
Thus for every vector \(f\),
For a bounded operator \(T\), \(\sigma (T)\) denotes the spectrum of \(T\). We need the following well-known lemma, see [2, Prop. 6, p. 16].
Lemma 2.1
Let \(A\) and \(B\) be bounded operators on a Hilbert space. For \(\lambda \ne 0\), \(\lambda \in \sigma (AB)\) if and only if \(\lambda \in \sigma (BA)\).
The following lemma has a straightforward proof, which is omitted.
Lemma 2.2
If \(P\) is an idempotent bounded operator on a Hilbert space and \(\lambda \ne 0,1\), then
3 Spectrum of \(\chi _{(-\tau ,\tau )}+ S_\omega \)
In this section, we describe the spectrum of the operator \(\chi _{(-\tau ,\tau )}+ S_\omega \) on \({\varvec{L}}^2 (\mathbb {R})\), where \(\tau ,\omega > 0\).
For a fixed \(c > 0\), the integral operator on \({\varvec{L}}^2(-1,1)\) with the kernel
has eigenvalues \(\lambda _0 > \lambda _1 > \dots > 0\) [7, 8]. The eigenfunctions are the PSWFs, and the eigenvalues obey certain asymptotic formulas. We are only interested in the largest eigenvalue \(\lambda _0\), which has the following asymptotics [8]:
when \(c \rightarrow \infty \).
We show that the eigenvalues of \(T = \chi _{(-\tau ,\tau )}+ S_\omega \), acting on \({\varvec{L}}^2 (\mathbb {R})\), can be expressed in terms of those of the operator with kernel (13), acting not on \({\varvec{L}}^2 (\mathbb {R})\), but rather on \({\varvec{L}}^2(-\tau ,\tau )\).
To indicate the dependence on \(c\), we write \(\lambda _n(c)\).
Theorem 3.1
Fix \(\tau ,\omega > 0\) and let \(T = \chi _{(-\tau ,\tau )}+ S_\omega \). If \(\lambda \in \sigma (T)\) and \(\lambda \ne 0,1\), then
for some \(n\).
Proof
In this proof we can assume that \(\tau = 1\). The general case follows by a linear change of variables.
It follows from the assumptions that \(\lambda I - \chi _{(-1,1)}- S_\omega \) is singular, and so is the operator
where we used Lemma 2.2 for \((\lambda I - S_\omega )^{-1}\). Thus
We use Lemma 2.1 for the operators \(\chi _{(-1,1)}\) and \(I + \frac{1}{\lambda - 1} S_\omega \chi _{(-1,1)}\) and the assumption that \(\lambda \ne 0\), to conclude that
It follows that the operator
is singular, and so is
The operator \(\chi _{(-1,1)}S_\omega \chi _{(-1,1)}\) is Hilbert-Schmidt, and therefore compact, and so is \((\lambda I - \chi _{(-1,1)})^{-1} \chi _{(-1,1)}S_\omega \chi _{(-1,1)}\).
From the spectral theory of compact operators, it follows that \(1\) is an eigenvalue of
i.e., there is an \(f \in {\varvec{L}}^2 (\mathbb {R})\), \(f \ne 0\) such that
or, equivalently
Using (32), we note that \(\chi _{(-1,1)}f \ne 0\) in \({\varvec{L}}^2 (\mathbb {R})\), because \(\lambda f \ne 0\) in \({\varvec{L}}^2 (\mathbb {R})\). We now multiply (32) by \(\chi _{(-1,1)}\) on the left to obtain
and, consequently,
Thus \(\chi _{(-1,1)}f\) is an eigenfunction of the operator with kernel \(\frac{\sin \omega (x-y)}{\pi (x-y)}\) on \({\varvec{L}}^2(-1,1)\) with eigenvalue \((\lambda - 1)^2\).
Thus
for some \(n\). \(\square \)
Remark 3.2
Only Theorem 3.1 is used in the proof of Theorem 1.2. However, we devote the rest of this section to a complete description of the spectrum of the operator \(T = \chi _{(-\tau ,\tau )}+ S_\omega \).
In this proof we can again assume that \(\tau = 1\). The general case follows by a linear change of variables.
We can show that if \((\lambda - 1)^2 = \lambda _n (\omega )\), then there exists an eigenfunction for \(T\) with eigenvalue \(\lambda \).
It follows from Slepian’s theory that \(0 < \lambda _n (\omega ) < 1\). We write
where \(\psi _n\) is the \(n\)th PSWF and \(\widetilde{\psi }_n\) is the extension of \(\psi _n\) to \(\mathbb {R}\), i.e.,
and
We multiply (36) by \(\chi _{(-1,1)}\) to get
Moreover,
where we used the assumption that \(\lambda _n = (\lambda - 1)^2\).
Finally,
or, equivalently,
Thus all numbers of the form \(1 \pm \sqrt{\lambda _n (\omega )}, n = 0,1,\dots \) are eigenvalues of \(T\).
The point \(\lambda = 1\) is also in the spectrum \(\sigma (T)\), as an accumulation point of the eigenvalues.
It remains to consider \(\lambda = 0\). To prove that \(T\) is singular, we consider the sequence of functions
It is clear that
but
in \({\varvec{L}}^2 (\mathbb {R})\).
Thus we have shown that \(\sigma (T)\) consists of the eigenvalues of \(T\) of the form \(1 \pm \sqrt{\lambda _n (\omega )}, n = 0,1,\dots \), and the two additional points \(\lambda = 0\) and \(\lambda = 1\).
The spectrum of the sum of two orthogonal projections was described in [1] in a somewhat different setting.
4 Proof of the Theorem 1.2
In this section, we present the proof of Theorem 1.2.
Proof of Theorem 1.2
In this proof, we assume that
The general case follows by a linear change of variables.
For a fixed \(\tau > 0\), we consider the restriction \(\chi _{(-\tau ,\tau )}f\) of \(f\) to the interval \((-\tau ,\tau )\). The decay of \(f\) at infinity in (3) gives an estimate on \(f - \chi _{(-\tau ,\tau )}f\) in the \({\varvec{L}}^2\)-norm. Specifically,
Similarly, (4) implies that for a fixed \(\omega > 0\),
Setting \(\tau = \omega \), using (18), and combining (49), (54) and (55), yields
The operator \(T' = 2I - \chi _{(-\tau ,\tau )}- S_\omega \) is Hermitian. According to Theorem 3.1, its smallest eigenvalue \(\lambda _{min}\) satisfies
Consequently,
The eigenvalue \(\lambda _0\) satisfies (21). Thus, since \(c = \omega \tau = \omega ^2\), we obtain
We recall the elementary formula
Substituting (64) and (65) into (63) we obtain
Letting \(\omega \rightarrow \infty \), we deduce that \(\Vert f \Vert = 0\). \(\square \)
4.1 Alternative Proof
A reviewer of this paper has remarked that an alternative proof is possible based on the following result proved in [6, p. 68].
Theorem 4.1
If \(\Vert f \Vert = 1\),
then
We present an outline of an alternative proof of Theorem 1.2. Let us assume that \(\Vert f \Vert = 1\). It follows from (49) and (54) that
Consequently, for all sufficiently large \(\tau \)’s,
and
We recall that as \(x \rightarrow 0^+\),
Combining (72) and (73), we conclude that for every sufficiently large \(\tau \),
Similarly, for every sufficiently large \(\omega \),
Combining (74) and (75), and setting \(\tau = \omega \), we obtain
Setting \(\Omega = \omega \) and \(T = 2\omega \) in (69), we obtain
Consequently,
Substituting (64) into (65), we obtain
Substituting (79) into (73), we obtain
Combining (78) and (80), we arrive at the contradiction
Our proof of Theorem 1.2 uses some techniques similar to those in [6], e.g., a linear combination of the time and the frequency limiting operators is already considered in [6, equation (6)].
References
Bjørstad, P.E., Mandel, J.: On the spectra of sums of orthogonal projections with applications to parallel computing. BIT Num. Math. 31(1), 76–88 (1991)
Bonsall, F., Duncan, J.: Complete Normed Algebras. Springer-Verlag, New York (1973)
Cowling, M., Escauriaza, L., Kenig, C.E., Ponce, G., Vega, L.: The Hardy uncertainty principle revisited. Indiana Univ. Math. J. 59, 2007–2026 (2010)
Escauriaza, L., Kenig, C.E., Ponce, G., Vega, L.: The sharp Hardy uncertainty principle for Schrödinger evolutions. Duke Math. J. 155(1), 163–187 (2010)
Hardy, G.H.: A theorem concerning Fourier transforms. J. London Math. Soc. 8(3), 227–231 (1933)
Landau, H.J., Pollak, H.O.: Prolate spheroidal wave functions, Fourier analysis and uncertainty - II. Bell Syst. Tech. J. 40, 65–84 (1961)
Slepian, D.: Some comments on Fourier analysis, uncertainty and modeling. SIAM Rev. 25, 379–393 (1983)
Slepian, D., Sonnenblick, E.: Eigenvalues associated with prolate spheroidal wave functions of zero order. Bell Syst. Tech. J. 44, 1745–1758 (1965)
Tao, T.: Hardys uncertainty principle (2009). http://terrytao.wordpress.com/2009/02/18/hardys-uncertainty-principle. Accessed 25 March 2012
Acknowledgments
The authors thank the reviewers for their helpful comments and suggestions, which have greatly improved this paper. The authors are supported by the FWF Grants S10602-N13 and P23902-N13.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Chris Heil.
Rights and permissions
About this article
Cite this article
Pauwels, E., de Gosson, M. On the Prolate Spheroidal Wave Functions and Hardy’s Uncertainty Principle. J Fourier Anal Appl 20, 566–576 (2014). https://doi.org/10.1007/s00041-014-9319-4
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00041-014-9319-4