The uncertainty principle on Cayley graphs (Q1306204)

This interesting paper investigates graph theoretic analogues of various results from analysis such as the fundamental theorem of calculus and versions of the uncertainty principle which are analogues of that given for periodic functions by \textit{A. Grünbaum} [Trying to beat Heisenberg, Lect. Notes Pure Appl. Math. 122, 657-665 (1990; Zbl 0702.35199)]. The classical uncertainty inequality follows from the following inequality for Schwartz functions \(f(x)\) on \(\mathbb{R}\): \[ \int |xf(x)|^2dx\int |f'(x)|^2dx\geq 1/4\Biggl( \int |f(x)|^2dx\Biggr) ^2 \] [see \textit{H. Dym} and \textit{H. P. McKean}, Fourier series and integrals (New York 1972; Zbl 0242.42001)]. If \(X\) is a finite oriented graph, consider a parameterization function \(p:X\rightarrow \mathbb{Z}\) such that \(p(x)\geq 0\). This parameterization function replaces the function \(p(x)=x\) in the uncertainty inequality. The standard parameterization function for a path \(C=e_1\cdots e_n\) in \(X\) with \(n\) edges \(e_i\) will be \(p(e_i)=i\), defining \(p\) to be \(0\) off \(C\). If \(f:X\rightarrow \mathbb{R}\), define \(df(e)=f(e^{+})-f(e^{-})\) for an edge \(e\) from vertex \(e^{+}\) to \(e^{-}\) of \(X\). \textbf{Integration by parts} on path \(C=e_1\cdots e_n\) says for \(f,g:X\rightarrow \mathbb{R}\), \[ \sum_{e\in C}df(e)g(e^{+})=[f(a)g(a)]_{e_1^{-}}^{e_n^{+}}-\sum_{e\in C}dg(e)f(e^{-}). \] Suppose \(f:X\rightarrow \mathbb{R}\), \(f(e_n^{+})=f(e_1^{-})=0\) and \(f\) vanishes off the vertices of \(C\). Define the \textbf{average function on edges} by \[ Av(f)(e)=\frac{f(e^{+})+f(e^{-})}2. \] The \textbf{uncertainty inequality} for path \(C\) proved in the last section of the paper says, if \(f\) is as above and \((dp)(e)\geq 0\) for all edges \(e\) (or \((dp)(e)\leq 0\) for all \(e\)): \[ \sum_{e\in C}( p(e^{+})Av(f)(e)) ^2\sum_{e\in C}( df(e)) ^2\geq \frac 14\Bigl( \sum_{e\in C}f(e^{-})^2|dp(e)|\Bigr) ^2. \] The proof proceeds as for the classical result beginning using integration by parts to see that \[ \sum_{e\in C}df(e)p(e^{+})\left( f(e^{+})+f(e^{-})\right) =-\sum_{e\in C}f(e^{-})^2dp(e). \] The Cauchy-Schwarz inequality finishes the proof. The paper (Theorem 3) gives a 2nd version of the uncertainty inequality. For this, one needs to assume \((dp)(e)\leq 0\), for all edges \(e\). Then \(Av(f)(e)\) is replaced with \(f(e^{+})\) on the left-hand side of the inequality. The paper has a typo in the hypothesis of Theorem 3, flipping the inequality for \((dp)(e)\). The paper also contains a version of uncertainty for the finite circle which is compared with that for the continuous circle given by Grünbaum in the paper mentioned above.