Square Integrable Group Representations and the Uncertainty Principle

Abstract

Let U be a square integrable representation of a Lie group G of transformations in a Hilbert space H, and let ψ ∈ H be an admissible state. We call the product of variances in the state ψ, associated to two non-commuting infinitesimal operators T 1 and T 2 , uncertainty measure.We investigate, in this note, how uncertainty measures of admissible states ψ are changing when ψ is transformed according to U(g)ψ for g ∈ G. We derive these transform laws for certain types of the associated Lie-algebra and apply them to relevant transform groups (affine group, similitude group, shearlets), investigated recently in signal processing.By doing so, we are able to show that the associated uncertainty measures can be made arbitrarily small in many cases. This implies that the corresponding "uncertainty minimizers", which are traditionally defined via equality in the famous uncertainty inequality and which have been investigated recently for above mentioned groups, actually do not minimize the uncertainty measure.In cases, where the uncertainty measure of U(g)ψ cannot be made arbitrarily small, when g varies over G, the transform laws derived in this paper indicate how to pick group elements g such that the uncertainty measure of U(g )ψ attains a local minimum. This method potentially enables discretizations of square integrable group representations with "low uncertainty", or equivalently "high locality" of the features of interest. This is of great importance for connecting uncertainty principles with frame constructions, and for maximum accuracy of feature measurements in signals and images.Communicated by Stephan Dahlke.

Uncertainty principles and optimally sparse wavelet transforms

“…These optimal window functions were never applied in engineering. Indeed the results are quite strange and counter intuitive, e.g [33]. Our assertion is that the conventional generalization is flawed, in the sense that the derived localization notions do not correspond to the "metaphysical concept" of localization.…”

Wavelet Design with Optimally Localized Ambiguity Function: a Variational Approach

“…Even after applying variational methods to find the minimizer of the left-and-side of (1), the above approach yields rather strange and counter-intuitive results (see e.g., [24]). The reason for that, as suggested in [20], is that the group generators T j are not appropriate for defining localization of the physical quantities underlying each transformation group e iRTj .…”

“…Note that the third and fourth equalities in (24) follows from linearity, the definition (7) of inner product of simple vectors in W ⊗ S, and the fact that ln(ω…”