2

Let $y\geq x\geq 1$ and $k \in \mathbb{N} \geq 1$. I want to find $f_{x,y} \geq \dfrac{1}{x^k}$ such that:

  1. $f(x,x) = \dfrac{1}{x^k}$ (This condition is to avoid trivial answers like $ y^2 \geq \dfrac{1}{x}$ for instance.)

This condition gives the general form for $f(x,y)$ which is $\sum a_i x^{b_i} y^{c_i} $ such that $b_i+c_i=-k$.

  1. $\sum$ $(\text{coefficients} \cdot (-\text{powers of } x)) < k$

If we took this form $f(x,y)=\sum a_i x^{b_i} y^{c_i}$ then the condition becomes $ \sum a_i · (-b_i) < k$.

For instance with $k=4$ and $f(x,y) = \dfrac{y}{x^5} \geq \dfrac{1}{x^4}$, the first condition holds but the second doesn't because $\text{coefficients}=1$ and $\text{power} = 5$ and so the total sum is $1\cdot 5= 5 >4$.

To find the sum in condition 2 for $f(x,y)= \dfrac{1}{x^2 y^2}+\dfrac{2}{x^3 y}$, we calculate $ 1\cdot 2+2\cdot 3 = 8 >4$.

I conjecture that there is no such function but I couldn't prove it.

Ѕᴀᴀᴅ
  • 34,263
Ahmad
  • 1,380
  • 1
  • 18
  • 37
  • 2
    This question needs clarity. It seems to assume the function has some specific form, and condition $(2)$ is not at all clear. – Fimpellizzeri Jun 23 '21 at 20:34
  • @Fimpellizieri I will edit it more to clarify, i did explain how to calculate the value of the second condition. – Ahmad Jun 23 '21 at 20:46
  • You mix the notation "$f(x,y)$" and "$f_{x,y}$". Do subscripts denote differentiation or is "$f_{x,y}$" synonymous with "$f(x,y)$"? – Eric Towers Jun 23 '21 at 21:10
  • Condition $2$ seems unintelligible if $f$ is not a sum of monomials, for instance , $\mathrm{e}^x$ doesn't seem to have any interpretation relative to your expression "$\text{-powers}$". – Eric Towers Jun 23 '21 at 21:10
  • @EricTowers $f_{x,y} = f(x,y)$ not derivates. – Ahmad Jun 23 '21 at 21:38
  • @EricTowers because i am looking for $f$ as a sum of monomials, i edited the question you can see the form i am looking for. – Ahmad Jun 23 '21 at 21:39
  • $k, a_i, b_i, c_i$ restricted to complexes? reals? integers? positive integers? – Eric Towers Jun 23 '21 at 21:57
  • 1
    @EricTowers $k$ is positive integer and $ b_i+ c_i = k$ but they can be any real numbers, and so does $a_i$ – Ahmad Jun 23 '21 at 22:12
  • @Ahmad So $a_i,b_i,c_i$ can be negative? – Ѕᴀᴀᴅ Jun 24 '21 at 02:52
  • @Saad In every case at least one of $b_i , c_i$ is negative and in lots of cases both are, and $a_i$ can be any real number(negatives included). – Ahmad Jun 24 '21 at 02:54
  • @Ahmad For $f(x, y) = \sum a_i x^{b_i} y^{c_i}$, condition 1 means $\sum a_i = 1$ and $b_i + c_i = -k$? For $f(x, y)= \dfrac{1}{x^2 y^2}+\dfrac{2}{x^3 y}$, it does not hold $f(x, x) = \frac{1}{x^4}$? – River Li Jun 24 '21 at 02:55
  • @RiverLi true, because that would be "trivial answer" , also you did it for $k=4$, you can pick any positive value for $k$ not just $4$ (just sayin). – Ahmad Jun 24 '21 at 02:58
  • @Ahmad So, we may rephrase the question as follows (?): Let $k$ be a positive integer. Let $f(x, y)$ be a polynomial of the form $f(x, y) = \sum_{i=1}^N a_i x^{b_i}y^{c_i}$ where $a_i, b_i, c_i$'s are real numbers such that $b_i + c_i = -k$ for all $i$, and $\sum_i a_i = 1$, and $\sum_i (-a_i b_i) < k$. Then there exists $y \ge x \ge 1$ such that $f(x, y) < \frac{1}{x^k}$. – River Li Jun 24 '21 at 03:12
  • @RiverLi the last line is opposite of what i want (for all $y\geq x\geq 1$) we have that $f(x,y) \geq \frac{1}{x^k}$ , or you can prove that no such $f$ exists. – Ahmad Jun 24 '21 at 13:26
  • @Ahmad Thanks. @ Saad's formulation is better. – River Li Jun 24 '21 at 13:57

1 Answers1

2

$\def\N{\mathbb{N}}\def\R{\mathbb{R}}\def\paren#1{\left(#1\right)}$Lemma 1: Suppose $a_1, \cdots, a_n, u_1, \cdots, u_n \in \R$ and $\sum\limits_{k = 1}^n a_k = 1$. If$$ \sum_{k = 1}^n a_k t^{u_k} \geqslant 1,\quad \forall 0 < t \leqslant 1 $$ then $\sum\limits_{k = 1}^n a_k u_k \leqslant 0$.

Proof: Define $F(t) = \sum\limits_{k = 1}^n a_k t^{u_k}$. Note that $F'(1)$ exists and $F(1) = \sum\limits_{k = 1}^n a_k = 1$. Suppose $F'(1) > 0$, there exists $0 < δ < 1$ such that $F(t) < F(1) = 1$ for $t \in (1 - δ, 1)$, a contradiction. Thus $F'(1) \leqslant 0$, i.e.\begin{gather*} F'(1) = \left.\paren{ \sum_{k = 1}^n a_k u_k t^{u_k - 1} }\right|_{t = 1} = \sum_{k = 1}^n a_k u_k \leqslant 0. \tag*{$\Box$} \end{gather*}

Now return to the question. The proposition to be proved can be formalized as below (@RiverLi):

Proposition: Suppose $m, n \in \N_+$. For any $a_1, \cdots, a_n, b_1, \cdots, b_n, c_1, \cdots, c_n \in \R$, if $\sum\limits_{k = 1}^n a_k = 1$, $b_k + c_k = -m$ for $1 \leqslant k \leqslant n$, and\begin{gather*} \sum_{k = 1}^n a_k x^{b_k} y^{c_k} \geqslant \frac{1}{x^m},\quad \forall y \geqslant x \geqslant 1 \tag{1} \end{gather*} then $\sum\limits_{k = 1}^n a_k · (-b_k) \geqslant m$.

Proof: Since $b_k + c_k = -m$ for all $k$, then$$ x^m \sum_{k = 1}^n a_k x^{b_k} y^{c_k} = \sum_{k = 1}^n a_k x^{b_k + m} y^{c_k} = \sum_{k = 1}^n a_k \paren{ \frac{x}{y} }^{b_k + m}, $$ thus (1) is equivalent to$$ \sum_{k = 1}^n a_k t^{b_k + m} \geqslant 1.\quad \forall 0 < t \leqslant 1 $$ Therefore, lemma 1 implies that $\sum\limits_{k = 1}^n a_k (b_k + m) \leqslant 0$, then\begin{gather*} -\sum_{k = 1}^n a_k b_k \geqslant m \sum_{k = 1}^n a_k = m. \tag*{$\Box$} \end{gather*}


In fact, the condition in lemma 1 is also necessary when $a_1, \cdots, a_n > 0$.

Lemma 2: Suppose $a_1, \cdots, a_n > 0$, $u_1, \cdots, u_n \in \R$. If $\sum\limits_{k = 1}^n a_k = 1$ and $\sum\limits_{k = 1}^n a_k u_k \leqslant 0$, then$$ \sum_{k = 1}^n a_k t^{u_k} \geqslant 1.\quad \forall 0 < t \leqslant 1 $$

Proof: Because $\sum\limits_{k = 1}^n a_k = 1$, the AM-GM inequality implies that for $0 < t \leqslant 1$,\begin{gather*} \sum_{k = 1}^n a_k t^{u_k} \geqslant \prod_{k = 1}^n (t^{u_k})^{a_k} = t^{\sum\limits_{k = 1}^n a_k u_k} \geqslant 1. \tag*{$\Box$} \end{gather*}

Ѕᴀᴀᴅ
  • 34,263