3
$\begingroup$

It is known about numbers $$a_1, a_2, ... a_n$$ that $$a_1 + a_2 +...+a_n \le 1/2.$$ Prove that $$ (1-a_1)(1-a_2)...(1-a_n) \ge 1/2$$ I have tried using $a^2 \geq 0$, it led to nothing. How can I make my inequality look like in the possible duplicate?

  • 0
    Arithmetic Mean - Geometric Mean inequality (AM-GM). – 2017-01-06
  • 3
    @mathreadler How is the AM-GM mean work here if we're not given the numbers are non-negative? – 2017-01-06
  • 3
    Isn't that wrong for $a_1 = 1, a_2 = -1/2$ ? Or are all numbers assumed to be non-negative? – 2017-01-06
  • 0
    Yes Martin is right. It is wrong if we can not assume non-negativity. – 2017-01-06
  • 0
    The inequality seems wrong; consider $f(x)=(x-a_1)(x-a_2)....(x-a_n)$. It is ONLY given that the sum of roots is less than or equal to $\frac{1}{2}$.. using this information no conclusion can be made about the value/range of $f(1)=(1-a_1)(1-a_2)...(1-a_n)$ – 2017-01-06
  • 0
    For *non-negative* numbers you can easily prove by induction that $(1-a_1)(1-a_2)...(1-a_n) \ge 1 - (a_1 + a_2 +...+a_n)$. But that must have been answered before ... – 2017-01-06
  • 0
    Arnaldo: all numbers can't be lower than the arithmetic average – 2017-01-06
  • 0
    @mathreadler: Sure! My bad! – 2017-01-06
  • 0
    And can you advise me how to make it look like AM-GM? – 2017-01-06
  • 0
    Or how to make it look like that? (1−a1)(1−a2)...(1−an)≥1−(a1+a2+...+an) By the way, this is the Bernoulli's inequality – 2017-01-06
  • 0
    I am trying to check now what my trail of thoughts were. – 2017-01-06
  • 0
    See also [How to show $x_1,x_2, \dots ,x_n \geq 0 $ and $ x_1 + x_2 + \dots + x_n \leq \frac{1}{2} \implies (1-x_1)(1-x_2) \cdots (1-x_n) \geq \frac{1}{2}$](http://math.stackexchange.com/q/216721). Found [using Approach0](https://approach0.xyz/search/?q=%24(1-a_1)(1-a_2)...(1-a_n)%20%5Cge%201%2F2%24&p=1). – 2017-01-06

2 Answers 2

0

By induction for $n=1$ it is obivious. You can consider $b_n = a_n+a_{n+1}$ where one has $$a_1, a_2, ... a_n$$ that $$a_1 + a_2 +...+\underbrace{a_n +a_{n+1}}_{b_n}\le 1/2 .$$But $$(1-a_n)(1-a_{n+1}) = (1-a_n-a_{n+1} + a_n a_{n+1} )\geq (1-a_n-a_{n+1})= (1-b_n).$$

Then $$(1-a_1)(1-a_2)...(1-a_n)(1-a_{n+1})\ge (1-a_1)(1-a_2)...(1-b_n) \ge 1/2$$

  • 0
    See @Martin's comment. – 2017-01-06
  • 0
    Presumably it is better to then put $0\leq a_i$ – 2017-01-06
-3

Hint: you can consider the inequality for arithmetic and geometric means:

$$\frac{1}{n}\sum_{k=1}^n b_i \geq \sqrt[n]{\prod_{k=1}^n b_i}$$

for any set of non-negative real numbers $\{b_1,\cdots,b_n\}$

Partial attempt:

$$b_i = 1-a_i$$ $$\frac{1}{n}\sum_{k=1}^n (1-a_i) \geq \sqrt[n]{\prod_{k=1}^n (1-a_i)}$$ $$1-\frac{1}{n}\sum a_i \geq \sqrt[n]{\prod_{k=1}^n (1-a_i)}$$

$$lhs > 1-\frac{1}{2n}$$

if product $\leq 1/2$ then rhs not sure to be larger than $\sqrt[n]{\frac{1}{2}}$

Now what to remains is to compare which can be done by calculus comparing a linear function with a power function, treating n as continous variable.

enter image description here

The proof that blue does not overtake red can be left as an exercise.

  • 2
    Just to be clear: before giving your hint, you did check that it worked, right? (It may just be me, but I don't really see how this helps) – 2017-01-06
  • 0
    @ClementC Okay, I have given some sketch. Please point out anything unrealistic that you find. – 2017-01-06
  • 0
    I am about to head out, but quickly: since $\sum_i b_i = \sum_(1-a_i) = n - \sum_i a_i \geq n/2$, isn't the inequalty written "lhs" in the wrong direction? – 2017-01-06
  • 0
    the sum is subtracted, so something smaller than $\frac{1}{2n}$ is being subtracted. – 2017-01-06
  • 0
    No, the way it is written $\frac{1}{n}\sum_i b_i \geq 1-\frac{1}{2n}$, since you defined $b_i=1-a_i$ and $\frac{1}{n}\sum_i a_i \leq \frac{1}{2n}$. – 2017-01-06
  • 0
    Yes wait maybe you are right, the b:s should be a:s and vice versa. So I edited so the $b_i$ are in the inequality and being substituted for $1-a_i$ instead. – 2017-01-06
  • 0
    (From my phone) i don't see how the last bit of your suggested proof goes. Given the inequalities we have and want, why does the rhs have to be greater than $1/2^{1/n} $? – 2017-01-06
  • 0
    Monotonicity of exponential function together with the $1/2$ in the value for of the product. Elementary calculus. – 2017-01-06
  • 0
    I don't think this addresses what I didn't get. You have (1) $lhs \geq \sqrt[n]{\prod_{i=1}^n (1-a_i)}$ and (2) $lhs > 1-\frac{1}{2n}$. Why would $1-\frac{1}{2n} < \sqrt[n]{\frac{1}{2}}$ bring any contradiction? – 2017-01-06