Bases and Frames
A basis is a linearly independent set of vectors that spans a given Hilbert space such that any function in that space can be expanded as a unique combination of these linearly independent vectors.On the other hand, frames are a generalization of bases such that they may be linearly dependent and, therefore, redundant. Formal definitions follow.
Definition of Bases: Consider a set of vectors {ek}∞k=1 in H
- 1. This set is a basis in H if for each value of f∈H there is a set of unique scalars cfk such that f=∞∑k=1cfkek
- 2. The basis {ek}∞k=1 is an orthonormal basis if it is an orthonormal system <ek,ej>=δk,j
Orthonormal Basis properties
- 1. f=∑∞k=1<f,ek>ek , ∀f∈H
- 2. <f,g>=∑∞k=1<f,ek><ek,g> , ∀f,g∈H
- 3. ¯span{ek}∞k=1=H
- 4. ∑∞k=1|<f,ek>|2=‖, \forall f \in H
- 5. If \lt f,e_{k} \gt =0 then f = 0
Biorthogonal Bases
Definition: Sets of vectors \{\widetilde{e}_{k}\}_{k=1}^{\infty} and \{e_{k}\}_{k=1}^{\infty} constitute biorthogonal bases in H if
- 1. \forall k,j \in Z, \lt e_{k},\widetilde{e}_{j} \gt =\delta_{k,j}
- 2. For any f \in H there exist A,B \gt 0 such that A\|f\|^{2} \le \sum_{k}| \lt f,e_{k} \gt |^{2} \le B\|f\|^{2} \frac{1}{B}\|f\|^{2} \le \sum_{k}| \lt f,\widetilde{e}_{k} \gt |^{2} \le \frac{1}{A}\|f\|^{2}
Bases that satisfy the above two equations are called Riesz Bases. f \in H can be expanded as
f=\sum_{k=1}^{\infty} \lt f,e_{k} \gt \widetilde{e}_{k} = \sum_{k=1}^{\infty} \lt f,\widetilde{e}_{k} \gt e_{k}Limitations of Bases or why we need Frames
Bases are characterized by their expansion property, ie. any function f \in H can be expressed as a linear combination of basis vectors e_{k} that span H. However, it is possible to expand a function f \in H as a linear combination of another set of vectors \phi which may not be a basis in H. The representation in this case may be redundant but it may provide additional flexibility which may be a good given specific applications. Another problem with basis is that they may be difficult to construct and a small change in implementing one vector in the basis may destroy the basis so having extra vectors in a redundant expansion may not be such a bad idea. In other applications, as in image compression, we may not want to keep all the coefficients of signal expansion with a set of given vectors so it is quite possible that frames may be able to do the job instead of bases which need more tedious construction.
Frames
Definition: Consider a set of vectors \{\phi_{k}\}_{k=1}^{\infty} \in H and that there exist constants A,B \gt 0. \phi_{k} is a frame in H if
A\|f\|^{2} \le \sum_{k=1}^{\infty} | \lt f,\phi_{k} \gt |^{2} \le B\|f\|^{2}, \forall f \in HIf A=B then the frame is called a Tight Frame.
If vectors \{\phi_{k}\}_{k=1}^{\infty} \in H are linearly independent then the frame is non-redundant and is called a Riesz Basis.
Let \{\phi_{k}\}_{k=1}^{\infty} \in H be a frame with frame operator S and bounds A,B then following is true
- 1. S is bounded, invertible, adjoint and positive.
- 2. \{S^{-1}\phi_{k}\}_{k=1}^{\infty} \in H is a frame with frame operator S^{-1} and frame bounds A^{-1},B^{-1}.
- 3. If A,B are optimal frame bounds for \{\phi_{k}\}_{k=1}^{\infty} then A^{-1},B^{-1} are optimal frame bounds for \{S^{-1}\phi_{k}\}_{k=1}^{\infty}.
- 4. f=\sum_{k=1}^{\infty} \lt f,S^{-1}\phi_{k} \gt \phi_{k}, \forall f \in H and f=\sum_{k=1}^{\infty} \lt f,\phi_{k} \gt S^{-1}\phi_{k}, \forall f \in H Both converge for all f \in H.
- 1. f=\sum_{k=1}^{\infty} \lt f,\chi_{k} \gt \phi_{k}, \forall f \in H
- 2. f=\sum_{k=1}^{\infty} \lt f,\phi_{k} \gt \chi_{k}, \forall f \in H
- 3. \lt f,g \gt =\sum_{k=1}^{\infty} \lt f,\phi_{k}\gt \lt \chi_{k},g \gt, \forall f,g \in H
Frame Operators and Pseudo-Inverse : Let U be the frame operator associated with the frame \{\phi_{k}\}_{k=1}^{\infty} \in H and let U: H \rightarrow H be bounded with closed range then \{U\phi_{k}\}_{k=1}^{\infty} is a frame sequence with frame bounds A\|U^{\dagger}\|^{-2} and B\|U\|^{2}. U^{\dagger} is known as the pseudo inverse of the frame operator U and is defined as
U^{\dagger}=(U^{*}U)^{-1}U^{*}where U^{*} is the adjoint of U. As can be seen from the definition, U^{\dagger} is the left inverse.
Dual Frames: Assume that \{\phi_{k}\}_{k=1}^{\infty} \in H is an overcomplete frame then there exist frames \{\chi_{k}\}_{k=1}^{\infty} \in H for which f=\sum_{k=1}^{\infty} \lt f,\chi_{k} \gt \phi_{k}, \forall f \in H . Following holds for dual frames \phi_{k} and \chi_{k}