Spectral orders and convolution (Q1065320): Difference between revisions

Denote by \({\mathcal M}_+\) the set of all extended-real valued, nonnegative and measurable functions on \({\mathbb{R}}=(-\infty,\infty)\), equipped with Lebesgue measure m. To each function f in \({\mathcal M}_+\), assign its decreasing rearrangement \(f^*\) and symmetrical decreasing rearrangement \(f{\hat{\;}}\) defined by \(f{\hat{\;}}(t)=f^*(2| t|)\) for each \(t\in {\mathbb{R}}.\) The author studies the relation between symmetrical decreasing rearrangements and convolution product \((f\times g)(x)=\int^{\infty}_{-\infty}f(t)g(x-t)dt\) of f,g\(\in {\mathcal M}_+\) from viewpoints of spectral orders \(\prec \prec\) and \(\prec\) of Hardy, Littlewood and Pólya. Here, weak spectral inequality \(f\prec \prec g\) means that the value of integral of \(f^*\) over [0,s] is less than or equal to that of \(g^*\) over [0,s] for each \(s\in {\mathbb{R}}_+\), and strong spectral inequality \(f\prec g\) means that \(f\prec \prec g\) and both the values of integral of f and g over \({\mathbb{R}}\) are equal to each other, in the sense that both may be infinite. The author first proves that weak spectral inequality \(f_ 1\prec \prec f_ 2\) is equivalent to the following: \(f_ 1^{{\hat{\;}}}\times g_ 1^{{\hat{\;}}}\prec \prec f_ 2^{{\hat{\;}}}\times g_ 2^{{\hat{\;}}}\) whenever \(g_ 1\prec \prec g_ 2\). Further, he proves that the corresponding assertion is valid also for strong spectral inequality. The main theorem is the following, to which he gives an elementary and transparent proof based on its discretization: \(f\times g\prec f{\hat{\;}}\times g{\hat{\;}}\) for all f,g\(\in {\mathcal M}_+\). The author also shows that the above strong spectral inequality is equivalent to the following classical inequality due to F. Riesz: \[ \int^{\infty}_{- \infty}\int^{\infty}_{-\infty}f(x)g(y)h(x-y)dxdy\leq \int^{\infty}_{-\infty}\quad \int^{\infty}_{-\infty}f{\hat{\;}}(x)g{\hat{\;}}(y)h{\hat{\;}}(x-y)dxdy \] for any f,g and h in \({\mathcal M}_+\). Applying the main theorem, he obtains some new inequalities. One of the results states: suppose that \(f_ 1\), \(f_ 2\) and \(g_ 1\), \(g_ 2\) are probability density functions and that \(f_ 1\prec f_ 2\) and \(g_ 1\prec g_ 2\); then \(H(f_ 2^{{\hat{\;}}}\times g_ 1^{{\hat{\;}}})\leq H(f_ 1\times g_ 1)\), provided that the values of both sides are definite, where H(h)\(\equiv -\int^{\infty}_{- \infty}h\cdot \log hdm\) is the Shannon's differential entropy of a probability density function h.