Abstract
We study Daubechies’ time–frequency localization operator, which is characterized by a window and weight function. We consider a Gaussian window and a spherically symmetric weight as this choice yields explicit formulas for the eigenvalues, with the Hermite functions as the associated eigenfunctions. Inspired by the fractal uncertainty principle in the separate time–frequency representation, we define the n-iterate mid-third spherically symmetric Cantor set in the joint representation. For the n-iterate Cantor set, precise asymptotic estimates for the operator norm are then derived up to a multiplicative constant.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The problem of localizing signals in time and frequency is an old and important one in signal analysis. In applications, we often wish to analyze signals on different time–frequency domains, and we would therefore attempt to concentrate signals on said domains. Different approaches for how to construct such time–frequency localization operators have been suggested, either based on a separate or joint time–frequency representation of the signal (see [4, 13]). The localization operators, regardless of which we choose to work with, will however be limited by the fundamental barrier of time–frequency analysis, namely the uncertainty principles, which state that a signal cannot be highly localized simultaneously in both time and frequency. With regard to the localization operator, the limits posed by the uncertainty principles translate into the associated operator norm, as it measures the optimal efficiency of any given localization operator.
Many versions of the uncertainty principles exist (see [10]), and more recent versions start to take into account the geometry of the time–frequency domains. In particular, in [6], Dyatlov describes the development and applications of a fractal uncertainty principle (FUP) for the separate time–frequency representation, first introduced and developed in [3, 7, 8]. The relevant localization operator is the standard composition of projections \(\pi _{T}Q_{\Omega }\), where \(\pi _{T}\) and \(Q_{\Omega }\) project onto the sets T in time and \(\Omega \) in frequency, respectively. In the context of the FUP, the sets T and \(\Omega \) take the form of fractal sets. Here fractal sets are defined in terms of the general notion of \(\delta \)-regularity (see [6, Definition 2.2.]), as families of sets \(T(h), \Omega (h)\subseteq [0,1]\), dependent on a continuous parameter \(0<h\le 1\). The FUP is then formulated for this general class of sets when \(h\rightarrow 0\).
An illustrative example featured in [6] is the mid-third Cantor set, where both the time and frequency domain can be regarded as h-neighbourhoods of the Cantor set, say C(h). Notice that the FUP is originally framed such that the parameter h is also encoded in the Fourier transform \(\mathcal {F}_{h}\) (not unlike the normalization with Planck’s constant in quantum mechanics) such that \(\mathcal {F}_h f(\omega ) = \frac{1}{\sqrt{h}}\hat{f}(\omega h^{-1})\), where \(\hat{f}\) denotes the Fourier transform of f. If we instead formulate the FUP in terms of the regular Fourier transform, we now consider the family \(C(h)/\sqrt{h}\). Further, if we translate the continuous h into discrete iterations n, we obtain a sequence based on the n-iterate Cantor set, defined in increasing intervals depending on n. More precisely, if \(T=\Omega = C_n\) denotes the n-iterate defined in the interval [0, M], then the interval length satisfies
which means \(|C_n| \sim \left( {2}/{\sqrt{3}}\right) ^n \rightarrow \infty \) as \(n\rightarrow \infty \). However, by Theorem 2.13 in [6], there exist constants \(\alpha , \beta >0\) such that
We should expect some analogous result to the FUP when extending to the joint time–frequency representation (see Itinerary page 1 in [11]). Inspired by the Cantor set example in separate time and frequency, we derive a similar localization result for the Cantor set in the joint representation. In particular, we consider Daubechies’ localization operator, first introduced in [4], based on the Short-Time Fourier tranform, with a spherically symmetric weight function and a Gaussian window. The reason for these restrictions is that, as was shown in the aforementioned paper, we obtain explicit expressions for the eigenvalues of the localization operator, with the Hermite functions as the associated eigenfunctions.
The remainder of the paper is organized as follows: In Sect. 2 we provide a more detailed introduction to the Daubechies operator (Sects. 2.1, 2.2), in addition to some necessary results in the spherically symmetric context (Sect. 2.3). We also make clear what we mean by a spherically symmetric Cantor set (Sect. 2.4). New results are found in Sects. 3 and 4, which contains several estimates for Daubechies’ operator localizing on different spherically symmetric sets.
In particular, Sect. 3 contains two introductory examples of localization on spherically symmetric subsets, namely localization on a ring and a set of infinite measure. In Sect. 4, we finally consider the n-iterate spherically symmetric Cantor set. Here we derive precise asymptotic estimates (up to a multiplicative constant) for the operator norm (Theorem 4.2). A particular case of this two-parameter result, in terms of the radius R and iterate n, can be formulated as an estimate solely in terms of the parameter n or R. In the spherically symmetric context, we consider the condition
similar to condition (1.1). Hence, under the above condition, let \(P_{n}\) denote the Daubechies operator localizing on the n-iterate spherically symmetric Cantor set defined in the disk of radius \(R>0\). Then for some positive constants \(c_1 \le c_2\) the operator norm satisfies
This result is analogous to knowing the exponential \(\beta >0\) in (1.2) precisely.
2 Preliminaries
2.1 Fourier and Short-Time Fourier Transform
For a function \(f:\mathbb {R}\rightarrow \mathbb {C}\) the Fourier transform evaluated at point \(\omega \in \mathbb {R}\) is given by
If we interpret f as an amplitude signal depending on time, then its Fourier transform \(\hat{f}\) corresponds to a frequency representation of the signal. The pair \((f,\hat{f})\) does not, however, offer a joint description with respect to both frequency and time. For this purpose, we consider the Short-Time Fourier transform (STFT) (see Chapter 3 in [11]).
The STFT is often referred to as the “windowed Fourier transform” as this transform relies on an additional fixed, non-zero function, \(\phi :\mathbb {R}\rightarrow \mathbb {C}\), known as a window function. At point \((\omega ,t)\in \mathbb {R}\times \mathbb {R}\) the STFT of f with respect to the window \(\phi \) is then defined as
The transformed signal now depends on both time t and frequency \(\omega \), and we refer to the \((\omega ,t)\)-domain \(\mathbb {R}^2\) as the phase space or the time–frequency plane.
We will restrict our attention to signals and windows in \(L^2(\mathbb {R})\), which, by Cauchy–Schwarz’ inequality, implies that \(V_{\phi }f(\omega ,t)\) is well-defined for all points \((\omega ,t)\in \mathbb {R}^2\). Such restrictions also produce the following orthogonality relation
Equipped with the standard \(L^2\)-norms, we deduce that the STFT is a bounded linear map, with target space \(L^2(\mathbb {R}^2)\). If the window \(\phi \) is normalized, i.e. \(\Vert \phi \Vert _2 =1\), then the STFT becomes, in fact, an isometry onto some subspace of \(L^2(\mathbb {R}^2)\).
Further, by identity (2.1), the original signal f can be recovered from its phase space representation. Take any \(\gamma \in L^2(\mathbb {R})\) such that \(\langle \gamma ,\phi \rangle \ne 0\), then the orthogonal projection of f onto any \(g\in L^2(\mathbb {R})\) is given by
A canonical choice for \(\gamma \) is to set it equal to \(\phi \). Assuming \(\phi \) is normalized, these projections then read
Since any signal \(f \in L^2(\mathbb {R})\) is entirely determined by such inner products, the right-hand side of formula (2.2) provides a complete recovery from the STFT.
2.2 Daubechies’ Localization Operator
One approach for how to construct operators that localize a signal f in both time and frequency was suggested by Daubechies in [4]. These operators can be summarized as modifying the STFT of f by multiplication of a weight function, say \(F(\omega ,t)\), before recovering a time-dependent signal. The weight function aims at enhancing certain features of the phase space while diminishing others. Based on formula (2.2), we consider the sesquilinear functional \(\mathscr {P}_{F,\phi }\) on the product \(L^2(\mathbb {R})\times L^2(\mathbb {R})\), defined by
Assuming \(\mathscr {P}_{F,\phi }\) is a bounded functional, Riesz’ representation theorem ensures the existence of a bounded, linear operator \(P_{F,\phi }: L^2(\mathbb {R})\rightarrow L^2(\mathbb {R})\) such that
The operator \(P_{F,\phi }\) is our sought after time-frequency localization operator, which we will refer to as Daubechies’ localization operator. From the above definition, \(P_{F,\phi }\) is characterized by the choice of weight F and window function \(\phi \).
In particular, any real-valued, integrable weight F will produce self-adjoint, compact operators \(P_{F,\phi }\) whose eigenfunctions form a complete basis for the space \(L^2(\mathbb {R})\). Furthermore, the eigenvalues \(\{\lambda _k\}_k\) satisfies \(\sum _{k}|\lambda _k|\le \Vert F\Vert _{1}\), in addition to \(|\lambda _k|\le \Vert F\Vert _{\infty }\) for all k.
2.3 Spherically Symmetric Weight
For an arbitrary weight F and window \(\phi \) it remains a challenge to determine the eigenvalues of Daubechies’ localization operator \(P_{F,\phi }\). However, in [4], Daubechies narrows in her focus to operators with a normalized Gaussian window
and a spherically symmetric weight
where \(r^2 = \omega ^2+t^2\). For such operators, the Hermite functionsFootnote 1
are shown to constitute the eigenfunctions. Further, explicit expressions for the associated eigenvalues \(\{\lambda _k\}_k\) are derived.
Theorem 2.1
(Daubechies) Let \(P_{F,\phi }\) denote the localization operator with weight \(F(\omega , t) = \mathscr {F}(r^2)\) and window \(\phi \) equal to the normalized Gaussian in (2.3). Then the eigenvalues of \(P_{F,\phi }\) are given by
such that
where \(H_k\) denotes the k-th Hermite function.
Observe that the normalized Gaussian in (2.3) coincides with \(H_0\) in (2.5). It was shown recently in [2] that (for each j) the Hermite functions are also eigenfunctions of any localization operator with window \(H_j\) and a spherically symmetric weight. Nevertheless, we will always assume the window \(\phi \) to be the normalized Gaussian.
We will consider the case when F projects onto a spherically symmetric subset \(\mathcal {E}\subseteq \mathbb {R}^2\). This means \(\mathscr {F}\) equals the characteristic function of some subset \(E\subseteq \mathbb {R}_{+}\), i.e., \(\mathscr {F}(r) = \chi _{E}(r)\), such that \(\mathcal {E} = \{ (\omega ,t) \in \mathbb {R}^2 \ | \ \omega ^2+t^2 \in E\}\). As a matter of convenience, we will denote the associated Daubechies operator simply by \(P_{\mathcal {E}}\). By Theorem 2.1, the eigenvalue corresponding to the k-th Hermite function is then given by
where \(\pi \cdot E := \{ x \in \mathbb {R}_{+}\ | \ x\pi ^{-1} \in E\}\). Since the above integrands will appear frequently, we define, for simplicity, the functions
In Sect. 4 we require two basic properties of the integrands \(\{f_k\}_k\) (see Appendix A for additional details), namely
and
where E is some measurable subset of \(\mathbb {R}_{+}\).
2.4 Cantor Set
The mid-third Cantor set based in the interval [0, R] is constructed as follows: Start with the interval \(C_0(R)=[0,R]\). Each n-iterate \(C_n(R)\) is the union of \(2^n\) disjoint, closed intervals \(\{I_{j,n}\}_{j}\). To obtain the next iterate \(C_{n+1}(R)\) remove the open middle-third interval in every interval \(I_{j,n}\). Such iterations yield a nested sequence \(C_0\supseteq C_1 \supseteq C_2 \supseteq \dots \) The mid-third Cantor set C(R) on the interval [0, R] is then defined as the intersection of all the n-iterates, i.e.,
For each n-iterate, we define a corresponding map \(\mathcal {G}_{R,n}: \mathbb {R}\rightarrow [0,1]\) by
which we refer to as the n-iterate Cantor function. These functions will come into play in the latter part of Sect. 4, where we will utilize the fact that \(\{\mathcal {G}_{R,n}\}_n\) are all subadditive, i.e.,
which was shown by induction by Josef Doboš in [5].
In the spherically symmetric context, we consider the following Cantor set construction: For the disk of radius \(R>0\) centered at the orgin, we identify the n-iterate with the subset
This means we consider weights of the form
Based on formula (2.6), the eigenvalues of \(P_{\mathscr {C}_n(R)}\) can then be expressed as
3 Examples of Localization on Spherically Symmetric Sets
In this section we present estimates for the operator norm of Daubechies’ operator localizing on two different spherically symmetric sets. For this purpose, it would be sufficient to determine the largest eigenvalue of the operator and estimate said eigenvalue. Nonetheless, even with identity (2.6), it may prove difficult to determine which eigenvalue is the largest. Under such circumstances, we will instead attempt to derive a common upper bound for the eigenvalues.
3.1 Localization on a Ring: Asymptotic Estimate
The first example shows that any eigenvalue \(\lambda _k\) of Daubechies’ localization operator can, in principle, be the largest eigenvalue. Consider localization on a ring of inner radius \(R>0\) in phase space of measure 1, that is, consider the subset
with the associated localization operator \(P_{\mathcal {E}(R)}\). By (2.6), the eigenvalues of \(P_{\mathcal {E}(R)}\) become
Now, assume that \(\pi R^2 \in [m,m+1]\) for some \(m\in \mathbb {N}\cup \{0\}\). Since the difference \(f_k(r) - f_{k+1}(r)\) is negative precisely when \(r>k+1\), we obtain the ordering
and
Under these conditions, either \(\lambda _{m}(R)\) or \(\lambda _{m+1}(R)\) must be the largest eigenvalue. In particular, if \(\pi R^2 = m\), then \(\lambda _m(R)\) becomes the largest eigenvalue. In the next proposition we provide an estimate of the operator norm of \(P_{\mathcal {E}(R)}\).
Proposition 3.1
Let \(\mathcal {E}(R)\subseteq \mathbb {R}^2\) be as in (3.1). For any fixed \(\pi R^2 \ge 2\), there exists a positive, finite constant C such that the operator norm of \(P_{\mathcal {E}(R)}\) satisfies the bounds
Proof
Let \(n:= {\lfloor }\pi R^2{\rfloor }\), where \({\lfloor }\cdot {\rfloor }\) denotes the floor function, rounding down to the nearest integer. Apply a max–min-approximation of the integrands \(f_k(r)\) for \(r\in [n,n+2] \supseteq [\pi R^2, \pi R^2+1]\). In particular, \(f_n(n)\) serves as an upper bound and, by inequality (2.7), \(f_{n+1}(n)\) serves as a lower bound for the operator norm. That is,
Once we combine this with Stirling’s approximation formula for the factorial
we obtain
Expressing the above inequality in terms of R, we use that \(\pi R^2 -1 \le n \le \pi R^2\) and factor out \(1/\sqrt{\pi R^2}\). The error terms \(\pm C R^{-3}\), follows by Taylor expansion of the remaining factors about \(1/(\pi R^2) = 0\). \(\square \)
Remark
A careful reading of the Taylor series expansion reveals that for \(\pi R^2\ge 2\), the inequalities in Proposition 3.1 hold for constant \(C=\pi ^{-2}\).
3.2 Localization on Set of Infinite Measure
Next, we consider a non-trivial example of localization on a spherically symmetric set of infinite measure (see [12] for a similar example in the separate time-frequency representation). Define the subset
which we can identify as an infinite number of equidistant intervals in \(\mathbb {R}_{+}\). Although the above set has infinite measure, we maintain good control over the operator norm of \(P_{\mathcal {E}(s)}\) and can produce precise estimates in terms of the parameter s.
Theorem 3.1
Let \(\mathcal {E}(s)\subseteq \mathbb {R}^{2}\) be as in (3.2) with \(s\in [0,1]\). Then the operator norm of \(P_{\mathcal {E}(s)}\) satisfies the bounds
Further, there exists \(s_0>0\) such that
Proof
By formula (2.6), the eigenvalues read
For each integral over \([n,n+s]\), consider the maximum of \(f_k(r)\) for \(r\in [n,n+1]\) such that
and
We now claim that the following inequality holds
For \(k=1\), inequality (3.7) is verified by computing the series explicitly. While for \(k>1\), compare the series with the integral over \(\mathbb {R}_{+}\), that is
Thus,
Since \(1+4e^{-2} < C\), claim (3.7) follows. Combining results (3.5)–(3.7) yields the upper bound in (3.3). In the lower bound case of (3.3), it is sufficient to observe
For the equality case (3.4), note that inequality (3.7) ensures that there exists a constant \(0<C_0<C\) such that \(\lambda _k(s) \le C_0s\) for any \(k,s>0\). Since \((1-e^{-s})s^{-1}\rightarrow 1\) from below as \(s \rightarrow 0\), it follows that some \(s_0>0\) with property (3.4) exists. \(\square \)
Remark
In [1] Theorem 3, a more general localization result is presented for signals \(f\in M^{p}(\mathbb {R})\) with \(p\ge 1\). The result is similar as it provides an upper bound when localizing on sparse sets in phase space. Applied to signals \(f\in M^2 = L^2\) and the subset \(\mathcal {E}(s)\), Theorem 3 yields a somewhat coarser upper bound, namely \(\Vert P_{\mathcal {E}(s)}\Vert _{\text {op}}\le C' \sqrt{s}\) for some constant \(C'>0\).
4 Localization on Spherically Symmetric Cantor Set
In this section we consider localization on the n-iterate spherically symmetric Cantor set, i.e., the set \(\mathscr {C}_n(R)\) in (2.11). Hence, we consider the localization operator \(P_{\mathscr {C}_n(R)}\) and attempt to estimate its operator norm. Results are formulated in Sect. 4.1, with the proof strategy and formal proofs in the subsequent Sects. 4.2–4.4.
4.1 Results: Bounds for the Operator Norm
Below two theorems regarding the operator norm of \(P_{\mathscr {C}_n(R)}\) are presented. The first theorem shows to what extent the operator norm is bounded by the first eigenvalue \(\lambda _0(\mathscr {C}_n(R))\).
Theorem 4.1
The operator norm of \(P_{\mathscr {C}_n(R)}\) is bounded from above by
The second theorem is a precise asymptotic estimate of the operator norm of \(P_{\mathscr {C}_{n}(R)}\) (up to a multiplicative constant) based on the same asymptotic estimate for \(\lambda _0(\mathscr {C}_n(R))\).
Theorem 4.2
There exist positive, finite constants \(c_1 \le c_2\) such that for each \(n=0,1,2,\dots \)
Proofs of Theorem 4.1 and 4.2 can be found in Sects. 4.3 and 4.4 , respectively. At the end of Sect. 4.4, we also retrieve numerical estimates for the constants in Theorem 4.2.
If we now enforce condition (1.3), and note that \(2^n \sim (\pi R^2)^{2\frac{\ln 2}{\ln 3}}\), we obtain the following corollary:
Corollary 4.1
Suppose that the radius R depends on the iterate n such that \(\pi R^2 \sim 3^{\frac{n}{2}}\). Then there exists positive, finite constants \(c_1 \le c_2\) such that
Note that the above corollary is the same as result (1.4), except that we have expressed the inequality in terms of the radius R rather than the iterate n. On this form we have been able to express bounds for the operator norm in terms of quantity \(\delta = \frac{\ln 2}{\ln 3}\), which is the fractal dimension of the mid-third Cantor set. It would therefore be interesting to investigate whether the same or a similar statement holds when localizing on different Cantor sets, with a different fractal dimension \(0<\delta <1\).
4.2 Main Strategy: Relative Areas
Both theorems are obtained from the integral formula (2.12) for the eigenvalues \(\{\lambda _k(\mathscr {C}_n(R))\}_k\). However, as the number of intervals in the n-iterate Cantor set grows as \(2^n\), it soon becomes rather impractical to evaluate these integrals directly. Instead we consider the local effect on the integrals when increasing from one iterate to the next. In particular, this means we initially consider the integral of \(f_k\) over a single interval, say \([s,s+3L]\) for \(s\ge 0\) and \(L>0\). Then we attempt to determine the relative area left under the curve \(f_k\) once the mid-third of the interval is removed, i.e., we wish to understand the function
Computing the above integrals, \(\mathcal {A}_k(s,3L)\) can alternatively be expressed
Observe that \(\mathcal {A}_k(s,3L)\) is independent of the starting point s precisely when \(k=0\). In particular,
For this reason, calculations with regard to \(\lambda _0(\mathscr {C}_n(R))\) are significantly simpler than for the remaining eigenvalues. In particular, we have the recursive relation
which in return means
Ideally, if all the relative areas \(\mathcal {A}_k(s,L)\) were bounded by \(\mathcal {A}_0(L)\) regardless of starting point \(s>0\) and interval length \(L>0\), we would conclude that \(\lambda _0(\mathscr {C}_n(R))\) is always the largest eigenvalue. As it turns out, this is not the case, e.g.,
Nonetheless, in Sect. 4.3, we are able to determine a common bound for the eigenvalues in terms of \(\lambda _{0}(\mathscr {C}_n(R))\). Here, Lemma 4.4 is worth highlighting as it relies on the subadditivity of the Cantor function. Next, in Sect. 4.4, we compute the asymptotic estimates for \(\lambda _{0}(\mathscr {C}_n(R))\).
4.3 Proof of Theorem 4.1
We begin by comparing \(\mathcal {A}_k(s,3L)\) to \(\mathcal {A}_0(3L)\) when \(s\ge k\).
Lemma 4.3
Let \(\{\mathcal {A}_k\}_k\) be given by (4.1). Then
Proof
Consider the derivative of \(\mathcal {A}_k(s,L)\) with respect to s, which yields
where
By identity (4.2), it is clear that \(\lim _{s\rightarrow \infty }\mathcal {A}_k(s,3L) = \mathcal {A}_0(3L)\) for all \(L>0\). Thus, it suffices to show that \(N_k(s,L)\ge 0\) for all \(s\ge k\) and \(L>0\) for \(k=1,2,3,\dots \) Introduce the function
Then we may express \(N_k(s,L)\) as a single integral over [0, L] such that
Hence, the function \(N_k(s,L)\) is positive for all \(s\ge k\) if the derivative of \(\Phi (r,s,L)\) with respect to s is negative. Consider each of the square bracket terms \([\dots ]\) in definition (4.5) separately, that is
so that \(\Phi _k(r,s,L) = \Psi _k(r,s,L,0)+\Psi _k(r,s,L,L)\).
In order to easily evaluate the derivative of \(\Psi _k\), notice first that the arguments of \(f_k(\cdot )\) in each term of \(\Psi _k\) sum to a fixed value, namely
- (i)
\(2a := 2s + r + L + y\).
Further, introduce the corrections to each argument
- (ii)
\(\epsilon _1 := 2^{-1}(L-r-y)\) and \(\epsilon _2 := 2^{-1}(L+r-3y)\)
such that \(\Psi _k\) becomes
Since \(a=a(s)\) is the only quantity in the above expression that depends s, we obtain
By (i)–(ii) and since \(r\le L\), we always have the ordering
which means (4.6) is negative whenever the factor \((-a^2+\epsilon _1^2+ka)\) is negative. The latest claim is easily verified as \(|\epsilon _{1}|\le a-k\), and therefore \(\epsilon _{1}^2-(a-k)^2 = (ka-\epsilon _{1}^2-a^2)+k(a-k) \le 0\). Hence, for any \(y\in \{0,L\}, 0\le r\le L\) and \(\ s\ge k\), we conclude that
\(\square \)
By the latest lemma, any shifted n-iterate Cantor set \(C_{n}(\pi R^2)+s\) with \(s\ge k\) satisfies
which combined with (2.8) and then identity (4.4), yields
Next, we relate the integrals of \(f_k\) over the shifted n-iterates to the non-shifted n-iterates.
Lemma 4.4
Let \(L>0\). Then for every fixed \(k,n=0,1,2,\dots \), we have
- (A)
\(\int _{C_n(L)\cap [k,\infty [}f_k(r)\mathrm {d}r\le \int _{C_n(L)+k}f_k(r)\mathrm {d}r\ \) and
- (B)
\(\int _{C_n(L)\cap [0,k]}f_k(r)\mathrm {d}r\le \int _{C_n(L)+k}f_k(r)\mathrm {d}r\).
Proof
For case (A), since \(f_k(r)\) is monotonically decreasing for \(r>k\), it suffices to verify
In terms of the Cantor function \(\mathcal {G}_{L,n}\) in (2.9), the above claim reads
which is the same subadditivity property as in (2.10).
For case (B), consider the reflection of elements \(C_n(L)\cap [0,k]\) about the point k, that is, consider the subset
By (2.7), we have that
Similarly to (A), in order to prove (B), it suffices to show that
By definition (4.8), the set \(\mathcal {R}_{n,k}\) satisfies
We now apply the subadditivity of \(\mathcal {G}_{L,n}\) to \(\mathcal {G}_{L,n}(k) = \mathcal {G}_{L,n}((r-k) + (2k-r))\), from which claim (4.9) follows. \(\square \)
Now, combine inequality (4.7) with Lemma 4.4, to conclude
which is a restatement of Theorem 4.1.
4.4 Proof of Theorem 4.2
We formulate a precise asymptotic estimate for the first eigenvalue.
Proposition 4.1
There exist positive, finite constants \(a_1 \le a_2\) such that for each \(n=0,1,2,\dots \)
Proof
Combine the two identities (4.3), (4.4) to obtain
By the above identity, it is sufficient to show that
Exchange the product for a sum, and the above inequality is equivalent to
The above inequality can now be proven by means of the following two claims
- (i)
there exists a finite, positive constant\(\beta \)such that
$$\begin{aligned} 0\le \sum _{j=1}^{\infty }\left[ \ln \left( 1+y^{1/3^j}\right) -y^{1/3^j}\ln 2\right] \le \beta \ \textit{for } y\in [0,1], \ \textit{and} \end{aligned}$$ - (ii)
there exist finite constants\(\gamma _1\le \gamma _2\)such that
$$\begin{aligned} \gamma _1 \le \sum _{j=1}^{n}e^{-x/3^j}-\left( n-\frac{\ln (x+1)}{\ln 3}\right) \le \gamma _2 \ \textit{for } x \in [0,3^n]. \ \ \ \ \ \ \end{aligned}$$
For claim (i), consider the non-negative function \(g(y) := \ln (1+y)-y\ln 2\) for \(y\in [0,1]\). Since \(|g'(y)|\le g'(0) = 1-\ln 2\) for all \(y\in [0,1]\), g(y) can be bounded from above by the linear spline
Thus, the sum in claim (i) is bounded by \(\sum _{j=1}^{\infty }h(y^{1/3^{j}})\) for \(y\in [0,1]\). Since \(g(0) = h(0) = 0\), we may always assume that \(y>0\). Further, observe that for any \(0<y\le 1\), we have that \(y^{1/3^{j}}\nearrow 1\) as \(j\rightarrow \infty \). In particular, for any fixed \(0<y\le 1\), there exists a smallest\(j_0 \in \mathbb {N}\) such that \(y^{1/3^{j}}\ge 1/2\) for all \(j\ge j_{0}\). Based on our choice \(j_0\), we split the sum
and consider each sum separately. While the first sum is possibly empty, in the non-empty case, introduce the variable \(z_1:= y^{1/3^{j_0-1}}\in [0,1/2]\) such that
Similarly for the second sum, introduce the variable \(z_2:= y^{1/3^{j_0}}\in [1/2,1]\) such that
By direct comparison with the geometric series, that is, \(2^{-3^j}\le 2^{-j}\) and \(1-2^{-1/3^{j}} \le 3^{-j}\ln (2)\), both series \(\mathcal {S}_1\) and \(\mathcal {S}_2\) are convergent. Hence, claim (i) follows with \(\beta = g'(0)(\mathcal {S}_1+\mathcal {S}_2)\).
Claim (ii) is proven by similar means as (i), where we split the sum \(\sum _{j=1}^{n}e^{-x/3^j}\). In particular, for a fixed \(x\in [0,3^n]\), we split the sum at the point \(j_1:= \max \{{\lfloor }\frac{\ln (x)}{\ln (3)}{\rfloor }, 0\}\) such that \(\sum _{j=1}^{n}=\sum _{j=1}^{j_1}+\sum _{j=j_1+1}^{n}\). If the first sum is non-empty, set \(z_3 := e^{-x/3^{j_1}}\in [0,e^{-1}]\) such that
which is a convergent series. For the second sum, we utilize for \(y\ge 0\) the inequalities \(1-y \le e^{-y}\le 1\) to obtain lower and upper estimates. By comparison with the geometric series and since \(x/3^{j_1+1} \le 1\), we conclude that
Finally, by combining estimates (4.12)–(4.13) with the bounds \(\frac{\ln (x+1)}{\ln (3)}-1\le j_1 \le \frac{\ln (x+1)}{\ln (3)}\), claim (ii) follows with constants \(\gamma _1 = -3/2\) and \(\gamma _2 = 1+\mathcal {S}_3\). \(\square \)
Now, by applying the estimates of Proposition 4.1 with constants \(a_1\le a_2\) to Theorem 4.1, we obtain Theorem 4.2 with constants \(c_1 = a_1 \le c_2 = 2a_2\).
Remark
(Numerical estimates) From the proof of Proposition 4.1, we are also able to retrieve some numerical estimates for the constants \(a_1\le a_2\). It should, however, be noted that the method chosen in the proof is not meant to produce optimal constants. Nevertheless, with \(\mathcal {S}_1,\mathcal {S}_2,\mathcal {S}_3\) defined as in (4.10)–(4.12), we obtain the estimates
Notes
Due to the choice of normalization for the Fourier transform, both the Gaussian and the Hermite functions are normalized differently than in [4]. The normalization is chosen in accordance with Folland [9]. If \(h_k\) denotes the k-th Hermite function in [4], this relates to \(H_k\) in (2.5) by \(H_k(x) =\frac{2^{1/4}}{\sqrt{2^k k!}}h_k(\sqrt{2\pi }x)\).
References
Abreu, L. D., Speckbacher, MAbreu, L. D., Speckbacher, M.: Donoho–Logan large sieve principles for modulation and polyanalytic Fock spaces. arXiv:1808.02258 (2018)
Abreu, L.D., Gröchenig, K., Romero, J.L.: Harmonic analysis in phase space and finite Weyl–Heisenberg ensembles. J. Stat. Phys. 174, 1104–1136 (2019)
Bourgain, J., Dyatlov, S.: Spectral gaps without the pressure condition. Ann. Math. 187(3), 825–867 (2018)
Daubechies, I.: Time–frequency localization operators: a geometric phase space approach. IEEE Trans. Inf. Theory 34(4), 605–612 (1988)
Doboš, J.: The standard Cantor function is subadditive. Proc. Am. Math. Soc. 124(11), 3425–3426 (1996)
Dyatlov, S.: An introduction to fractal uncertainty principle. J. Math. Phys. 60, 081505 (2019)
Dyatlov, S., Jin, L.: Dolgopyat’s method and the fractal uncertainty principle. Anal. PDE 11(6), 1457–1485 (2018)
Dyatlov, S., Zahl, J.: Spectral gaps, additive energy, and a fractal uncertainty principle. Geom. Funct. Anal. 26, 1011–1094 (2016)
Folland, Gerald B.: Harmonic Analysis in Phase Space. vol. 122. Annals of Mathematics Studies. Princeton University Press, Princeton, N.J. (1989)
Folland, G., Sitaram, A.: The uncertainty principle: a mathematical survey. J. Fourier Anal. Appl. 3(3), 207–238 (1997)
Gröchenig, K.: Foundations of Time–Frequency Analysis. Applied and Numerical Harmonic Analysis. Birkhäuser, Boston (2001)
Reznikov, A.: Sharp constants in the Paneyah–Logvinenko–Sereda theorem. Comptes rendus - Mathématique 348(3), 141–144 (2010)
Slepian, D., Pollak, H.O.: Prolate spheroidal wave functions, Fourier analysis and uncertainty - I. Bell Syst. Tech. J. 40, 43–63 (1961)
Acknowledgements
Open Access funding provided by NTNU Norwegian University of Science and Technology (incl St. Olavs Hospital - Trondheim University Hospital). The author would like to thank Eugenia Malinnikova for many fruitful discussions on the topic covered and for providing feedback on drafts of the manuscript. In addition, the author would like to extend thanks to the anonymous referees for their many constructive comments on the first draft.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Elena Cordero.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The research of the author was supported by Grant 275113 of the Research Council of Norway.
Appendix A. Omitted Proofs in Sect. 2.3
Appendix A. Omitted Proofs in Sect. 2.3
We shall prove the following two properties for the integrands \(\{f_k(r):= \frac{r^k}{k!}e^{-r}\}_{k=0}^{\infty }\):
and
where E is some measurable subset of \(\mathbb {R}_{+}\).
Proof
(Property (A.1)) It is sufficient to show that the fraction \(\delta _k(r) := \frac{f_k(k-r)}{f_k(k+r)} \le 1\) for all \(r\in [0,k]\). By differentiation, \(\delta _k'(r) \le 0\) for all \(r\in [0,k]\) and since \(\delta _k(0) =1\), we are done. \(\square \)
Proof
(Property (A.2)) Since every \(f_k\) is normalized, i.e., \(\Vert f_k\Vert _1 = 1\), and \(f_k(r)\) is monotonically increasing for \(0<r<k\) and decreasing for \(r>k\), we may assume E to be an interval of finite measure. Define the function
and note that it suffices to show that \(g_k(L,s) \ge 0\) for all \(L,s\ge 0\) and every k. Differentiating \(g_k\) with respect to L,
reveals a critical point at \(L = L_0:= \root k \of {e^{s}k!} -s\). By the second derivative test, it follows that \(L=L_0\) represents a maximum for \(g_k(L,s)\) with \(s>0\) fixed. Since the other possible extrema occur when \(L=0\) or \(L\rightarrow \infty \), which both can easily be verified to yield a non-negative \(g_k(L,s)\), we conclude that \(g_k(L,s)\) is always non-negative. \(\square \)
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Knutsen, H. Daubechies’ Time–Frequency Localization Operator on Cantor Type Sets I. J Fourier Anal Appl 26, 47 (2020). https://doi.org/10.1007/s00041-020-09751-9
Received:
Revised:
Published:
DOI: https://doi.org/10.1007/s00041-020-09751-9