Home » Inequalities

Tag: Inequalities

Prove an inequality relating sums and integrals

Prove that

    \[ \sum_{k=1}^{n-1} f(k) \leq \int_1^n f(x) \, dx \leq \sum_{k=2}^n f(k) \]

where f is a nonnegative, increasing function defined for all x \geq 1.

Use this to deduce the inequalities

    \[ e n^n e^{-n} < n! < e n^{n+1} e^{-n} \]

by taking f(x) = \log x.


Proof. Since f(x) is increasing we know f([x]) \leq f(x) \leq f([x+1]) (where [x] denotes the greatest integer less than or equal to x) for all x \geq 1. Define step functions s and g by

    \[ s(x) = f([x]), \qquad t(x) = f([x+1]). \]

Then s and t are constant on the open subintervals of the partition

    \[ P = \{ 0,1,2, \ldots, n \}. \]

So,

    \begin{align*}  \int_1^n s(x) \, dx = \sum_{k=1}^{n-1} s_k (x_k - x_{k-1}) = \sum_{k=1}^{n-1} f(k) \\[9pt]  \int_1^n t(x) \, dx = \sum_{k=1}^{n-1} t_k (x_k - x_{k-1}) = \sum_{k=1}^{n-1} f(k+1) = \sum_{k=2}^n f(k). \end{align*}

Since f is integrable we must have

    \[ \int_1^n s(x) \, dx \leq \int_1^n f(x) \, dx \leq \int_1^n t(x) \, dx \]

for every pair of step functions s \leq f \leq t. Hence,

    \[ \sum_{k=1}^{n-1} f(k) \leq \int_1^n f(x) \, dx \leq \sum_{k=2}^n f(k). \qquad \blacksquare \]

Next, if we take f(x) = \log x (which is nonnegative and increasing on x \geq 1) we have

    \begin{align*}  &&\sum_{k=1}^{n-1} \log k \leq \int_1^n \log x \, dx \leq \sum_{k=2}^n \log k \\[9pt]  \implies && \log(n-1)! \leq n \log n - n + 1 \leq \log n! \\[9pt]  \implies && (n-1)! \leq e n^n e^{-n} \leq n!. \end{align*}

From this we then get two inequalities

    \[ n! \leq en^{n+1} e^{-n} \quad \text{and} \quad e n^n e^{-n} \leq n!. \]

Therefore,

    \[ en^n e^{-n} \leq n! \leq e n^{n+1} e^{-n}. \]

Prove some statements about integrals of bounded monotonic increasing functions

Consider a bounded, monotonic, real-valued function f on the interval [0,1]. The define sequences

    \[ s_n = \frac{1}{n} \sum_{k=0}^{n-1} f \left( \frac{n}{k} \right), \qquad t_n = \frac{1}{n} \sum_{k=1}^n f \left( \frac{k}{n} \right). \]

  1. Prove that

        \[ s_n \leq \int_0^1 f(x) \, dx \leq t_n \]

    and that

        \[ 0 \leq \int_0^1 f(x) \,dx - s_n \leq \frac{f(1) - f(0)}{n}. \]

  2. Prove that the two sequences \{ s_n \} and \{ t_n \} converge to \int_0^1 f(x) \, dx.
  3. State and prove a generalization of the above to interval [a,b].

  1. Proof.First, we define two step functions,

        \[ s(x) = f \left( \frac{[nx]}{n} \right), \qquad t(x) = f \left( \frac{[nx+1]}{n} \right) \]

    where [x] denotes the greatest integer less than or equal to x. Then we define a partition of [0,1],

        \[ P = \left\{ 0, \frac{1}{n}, \frac{2}{n}, \ldots, 1 \right\}. \]

    For any x_{k_1} \leq x < x_k we have

        \[ s(x) = f \left( \frac{k-1}{n} \right), \qquad t(x) = f \left( \frac{k}{n} \right). \]

    So, s(x) and t(x) are constant on the open subintervals of the partition P.
    Since f(x) is monotonically increasing and s(x) \leq f(x) \leq t(x) for all x (by the definition of s and t) we have

        \begin{align*}  &&\int_0^1 s(x) \, dx \leq \int_0^1 f(x) \, dx \leq \int_0^1 t(x) \, dx \\[9pt]  \implies && \sum_{k=0}^{n-1} s_k (x_k - x_{k-1} \leq \int_0^1 f(x) \, dx \leq \sum_{k=0}^{n-1} t_k (x_k - x_{k-1}) && (\text{Def step function integral}) \\[9pt]  \implies && \frac{1}{n} \sum_{k=0}^{n-1} f \left( \frac{k}{n} \right) \leq \int_0^1 f(x) \, dx \leq \frac{1}{n} \sum_{k=0}^{n-1} f \left( \frac{k+1}{n} \right) &&(x_k - x_{k-1} = \frac{1}{n} \forall k) \\[9pt]  \implies && \fract{1}{n} \sum_{k=0}^{n-1} f \left( \frac{k}{n} \right) \leq \int_0^1 f(x) \, dx \leq \frac{1}{n} \sum_{k=1}^nf \left( \frac{k}{n} \right) \\[9pt]  \implies && s_n \leq \int_0^1 f(x) \, dx \leq t_n && \text{for all } n \\[9pt]  \implies && 0 \leq \int_0^1 f(x) \, dx - s_n \leq t_n - s_n \\[9pt]  \implies && 0 \leq \int_0^1 f(x) \, dx - s_n \leq \frac{f(1) - f(0)}{n}. \qquad \blacksquare \end{align*}

  2. Proof. From part (a) we have

        \[ 0 \leq \int_0^1 f(x) \, dx - s_n \leq \frac{f(1) - f(0)}{n} \quad \implies \quad \lim_{n \to \infty} \left( \int_0^1 f(x) \, dx - s_n \right)= 0 \]

    since \lim_{n \to \infty} \frac{f(1)- f(0)}{n} = 0. Since \int_0^1 f(x) \, dx does not depend on n we have

        \[ \int_0^1 f(x) \, dx  - \lim_{n \to \infty} s_n = 0 \quad \implies \quad \lim_{n \to \infty} s_n = \int_0^1 f(x) \, dx. \]

    Therefore,

        \begin{align*}   && s_n \leq \int_0^1 f(x) \, dx \leq t_n \\[9pt]  \implies && s_n - t_n \leq \int_0^1 f(x) \, dx - t_n \leq 0 \\[9pt]  \implies && \frac{f(0)-f(1)}{n} \leq \int_0^1 f(x) \, dx - t_n \leq 0 \\[9pt]  \implies && \lim_{n \to \infty} \left( \int_0^1 f(x) \, dx - t_n \right) = 0 \\[9pt]  \implies && \lim_{n \to \infty} t_n = \int_0^1 f(x) \, dx. \qquad \blacksquare \end{align*}

  3. Claim: If f is a real-valued function that is monotonic increasing and bounded on the interval [a,b], then

        \[ \lim_{n \to \infty} s_n = \int_a^b f(x) \, dx, \qquad \text{and} \qquad \lim_{n \to \infty} t_n = \int_a^b f(x) \, dx \]

    for s_n and t_n defined as follows:

        \[ s_n = \frac{b-a}{n} \cdot sum_{k=0}^{n-1} f \left( a + k \frac{b-a}{n} \right), \qquad t_n = \frac{b-a}{n} \cdot \sum_{k=1}^n f \left( a + k \frac{b-a}{n} \right). \]

    Proof. Let

        \[ P = \left\{ a, a + \frac{b-a}{n}, a + \frac{2(b-a)}{n}, \ldots, a + \frac{n(b-a)}{n} = b \right\}  \]

    be a partition of the interval [a,b]. Then, define step functions s(x) and t(x) with s(x) = f(x_{k-1}) and t(x) = f(x_k) for x_{k-1} \leq x < x_k. By these definitions we have s(x) \leq f(x) \leq t(x) for all x \in [a,b] (since f is monotonic increasing). Since f is bounded and monotonic increasing it is integrable, and

        \begin{align*}  && \int_a^b s(x) \, dx \leq \int_a^b f(x) \, dx \leq \int_a^b t(x) \, dx \\[9pt]  \implies && \sum_{k=1}^n s_k (x_k - x_{k-1}) \leq \int_a^b f(x) \, dx \leq \sum_{k=1}^n t_k (x_k - x_{k-1}) \\[9pt]  \implies && \sum_{k=1}^n s_k \left( \frac{b-a}{n} \right) \leq \int_a^b f(x) \, dx \leq \sum_{k=1}^n t_k \left( \frac{b-a}{n} \right). \end{align*}

    And, since s_k = f(x_{k-1} = f \left( a + \frac{(k-1)(b-a)}{n} \right), and t_k = f(x_k) = f \left( a + \frac{k(b-a)}{n} \right) we have

        \begin{align*}  \implies && \sum_{k=1}^n f \left( a + \frac{(k-1)(b-a)}{n} \right) \left( \frac{b-a}{n} \right) \leq \int_a^b f(x) \, dx \leq \sum_{k=1}^n f \left( a + \frac{k(b-a)}{n} \right) \left( \frac{b-a}{n} \right) \\[9pt]  \implies && \left( \frac{b-a}{n} \right) \sum_{k=0}^{n-1} f \left( a+ \frac{k(b-a)}{n} \right) \leq \int_a^b f(x) \, dx \leq \left( \frac{b-a}{n} \right) \sum_{k=1}^n f \left( a+ \frac{k(b-a)}{n} \right). \qquad \blacksquare \end{align*}

Sketch inequalities in the complex plane

Sketch each of the following sets of complex numbers z that satisfy the given inequalities:

  1. |2z + 3| < 1.
  2. |z+1| < |z-1|.
  3. |z-i| \leq |z+i|.
  4. |z| \leq |2z+1|.

  1. Letting z = x+ iy we have,

        \begin{align*}  |2z+3| < 1 && \implies && |2x+3 + 2yi| &< 1 \\  && \implies && \sqrt{(2x+3)^2 + 4y^2} &< 1 \\  && \implies && (2x+3)^2 + 4y^2 < 1. \end{align*}

    This is a disk of radius \frac{1}{2} centered at \left(-\frac{3}{2}, 0 \right). The sketch is as follows:
    Complex4

  2. Letting z = x+ iy we have,

        \begin{align*}  |z+1| < |z-1| && \implies && \sqrt{(x+1)^2 +y^2} &< \sqrt{(x-1)^2 + y^2} \\  && \implies && x^2 + 2x + 1 + y^2 &< x^2 -2x + 1 + y^2 \\  && \implies && 4x &< 0 \\  && \implies && x &< 0. \end{align*}

    This is the half-plane with negative real part. The sketch is as follows:
    Complex5

  3. Letting z = x + iy we have,

        \begin{align*}  |z-i| \leq |z+i| && \implies && \sqrt{x^2 + (y-1)^2} &\leq \sqrt{x^2 + (y+1)^2} \\  && \implies && x^2 + y^2 - 2y + 1 &\leq x^2 + y^2 + 2y + 1 \\  && \implies && y & \geq 0. \end{align*}

    This is the half-plane with positive imaginary part. The sketch is as follows:
    Complex6

  4. Letting z = x+ iy we have,

        \begin{align*}  |z| \leq |2z+1| && \implies && \sqrt{x^2+y^2} &\leq \sqrt{(2x+1)^2 + (2y)^2} \\  && \implies && x^2 + y^2 &\leq 4x^2 + 4x + 1 + 4y^2 \\  && \implies && 0 &\leq 3x^2 + 4x + 1 + 3y^2 \\  && \implies && 0 &\leq x^2 + \frac{4}{3}x + \frac{4}{9} + y^2 - \frac{1}{9} \\  && \implies && \frac{1}{9} &\leq \left( x +\frac{2}{3} \right)^2 + y^2. \end{align*}

    This is the region outside the disk of radius \frac{1}{3} centered at the point \left( -\frac{2}{3}, 0 \right). The sketch is as follows:
    Complex7

Prove an inequality for the integral of 1 / (1 + x4)

Prove that

    \[ 0.493948 < \int_0^{\frac{1}{2}} \frac{1}{1+x^4} \, dx < 0.493958. \]

(Note: I cannot get the bounds Apostol asks for. I prove a different set below. I cannot figure out if it is a mistake in the book or not.)


Proof. Using the algebraic identity, valid for 0 < x < 1,

    \[ \frac{1}{1-x} = 1+x+x^2+\cdots + x^n + \frac{x^{n+1}}{1-x}, \]

we obtain

    \[ \frac{1}{1+x^4} = 1 - x^4 + x^8 - x^{12} + \cdots + (-1)^n x^{4n} + (-1)^{n+1} \frac{x^{4n+4}}{1+x^4}. \]

Therefore, integrating term by term,

    \begin{align*}  \int_0^{\frac{1}{2}} \frac{1}{1+x^4} \, dx &= \int_0^{\frac{1}{2}} \left( 1 - x^4 + x^8 + \cdots + (-1)^n x^{4n} + (-1)^{n+1} \frac{x^{4n+4}}{1+x^4} \right) \, dx \\[9pt]  &= \left( x - \frac{x^5}{5} + \frac{x^9}{9} - \cdots + (-1)^n \frac{x^{4n+1}}{4n+1} \right)\Bigr \rvert_0^{\frac{1}{2}} + (-1)^{n+1} \int_0^{\frac{1}{2}} \frac{x^{4n+4}}{1+x^4} \, dx. \end{align*}

Furthermore, we have

    \[ \int_0^{\frac{1}{2}} \frac{x^{4n+4}}{1+x^4} \, dx < \int_0^{\frac{1}{2}} x^{4n+4} \, dx = \left( \frac{1}{2} \right)^{4n+5} \left( \frac{1}{4n+5} \right). \]

Taking n = 2, we then have

    \[ \int_0^{\frac{1}{2}} \frac{1}{1+x^4} \, dx = \left( \frac{1}{2} \right) - \frac{(1/2)^5}{5} + \frac{(1/2)^9}{9} - \int_0^{\frac{1}{2}} \frac{x^{4n+4}}{1+x^4} \, dx.\]

From the inequality for this integral we then have

    \begin{align*} &\frac{1}{2} - \frac{(1/2)^5}{5} + \frac{(1/2)^9}{9} - \left( \frac{1}{2} \right)^{4n+5} \left( \frac{1}{4n+5} \right) < \int_0^{\frac{1}{2}} \frac{1}{1+x^4} \, dx\\ \intertext{and}  &\int_0^{\frac{1}{2}} \frac{1}{1+x^4} \, dx < \frac{1}{2} - \frac{(1/2)^5}{5} + \frac{(1/2)^9}{9} \\[9pt] \implies & 0.493958 < \int_0^{\frac{1}{2}} \frac{1}{1+x^4} \, dx < 0.49367. \qquad \blacksquare \end{align*}

Use Taylor polynomials to approximate the nonzero root of arctan x = x2

  1. Show that r = \frac{\sqrt{21}-3}{2} is an approximation of the nonzero root of the equation

        \[ \arctan x = x^2 \]

    using the cubic Taylor polynomial approximation to \arctan x.

  2. Given that

        \[ \sqrt{21} < 4.6 \qquad \text{and} \qquad 2^{16} = 65536 \]

    prove that the number r from part (a) satisfies

        \[ |r^2 - \arctan x| < \frac{7}{100}. \]

    Determine if (r^2 - \arctan r) is positive or negative and prove the result.


  1. Proof. From a previous exercise (Section 7.8, Exercise #3) we know

        \[ \arctan x = x - \frac{x^3}{3} + E_{2n}(x). \]

    So, to approximate the nonzero root of x^2 - \arctan x we have

        \begin{align*}  x^2 - x + \frac{x^3}{3} \approx 0 && \implies && x^2 + 3x - 3 &\approx 0 \\  && \implies && x &\approx \frac{\sqrt{21}-3}{2}. \qquad \blacksquare \end{align*}

  2. We know from the same previous exercise we used in part (a) that the error term E_{2n}(x) for \arctan x satisfies the inequality

        \[ |E_{2n}(x)| \leq \frac{x^{2n+1}}{2n+1}. \]

    Using the values for \sqrt{21} and 2^{16} given we have

        \begin{align*}  \left|E_5 \left( \frac{\sqrt{21}-3}{2} \right)\right| &\leq \left| \frac{\left( \frac{\sqrt{21}-3}{2} \right)^5}{5} \right|  \leq \left| \frac{0.8^5}{5} \right| = \frac{(4/5)^5}{5}  \\[9pt]  &= \frac{2^{10}}{5^6} = \frac{2^{16}}{10^6} = \frac{65536}{1000000} < \frac{7}{100}. \qquad \blaacksquare \end{align*}

Use Taylor polynomials to approximate the nonzero root of x2=sin x

  1. Using the cubic Taylor polynomial approximation of \sin x, show that the nonzero root of the equation

        \[ x^2 = \sin x \]

    is approximated by r = \sqrt{15} - 3.

  2. Using part (a) show that

        \[ | \sin r - r^2  | < \frac{1}{200}, \]

    given that \sqrt{15} - 3 < 0.9. Determine whether (\sin r - r^2) is positive or negative and prove the result.


  1. Proof. The cubic Taylor polynomial approximation of \sin x is

        \[ \sin x = x - \frac{x^3}{3!} + E_{2n} (x). \]

    This implies

        \[ x^2 - \sin x \approx x^2 - x + \frac{x^3}{6}. \]

    Therefore, we can approximate the nonzero root by

        \begin{align*}  x^2 - \sin x = 0 && \implies && x^2 - x + \frac{x^3}{6} &\approx 0 \\  && \implies && 6x^2 + x - 1 &\approx 0 \\  && \implies && x &\approx \sqrt{15} - 3.  \qquad \blacksquare \end{align*}

  2. Proof. We know from this exercise (Section 7.8, Exercise #1) that for \sin x we have

        \[ |E_{2n} (x)| \leq \frac{|x|^{2n+1}}{(2n+1)!}. \]

    So, for n = 2, and using the given inequality \sqrt{15} - 3 < 0.9, we have

        \[ |E_{2n} (x)| \leq \frac{|0.9|^5}{120} = \frac{9^5}{10^5 (120)} < \frac{1}{200}. \]

    Furthermore, (\sin r - r^2) > 0 since

        \[ \sin r - r^2 = E_{2n}(x) = \frac{x^5}{5!} - \frac{x^7}{7!}+ \cdots \]

    with the absolute value of each term in the sum strictly less than the absolute value of the previous term (since x < 1 and (n+2)! > n!). Thus, each pair is positive, so the whole series is positive. \qquad \blacksquare

Prove an inequality for the error of the Taylor polynomial of arctan x

Prove that the error term in the Taylor expansion of \arctan x satisfies the following inequality.

    \[ \arctan(x) = \sum_{k=0}^{n-1} \frac{(-1)^k x^{2k+1}}{2k+1} + E_{2n} (x), \qquad |E_{2n}(x)| \leq \frac{x^{2n+1}}{2n+1}, \quad \text{if } 0 \leq x \leq 1. \]


Proof. To prove this we will work directly from the definition of the error as an integral,

    \[ E_n (x) = \frac{1}{n!} \int_a^x (x-t)^n f^{(n+1)}(t) \, dt. \]

We know for 0 \leq x \leq 1 we have, (we need 0 \leq x \leq 1 for the expansion of \frac{1}{1+x^2} to be valid),

    \begin{align*}  \arctan x &= \int_0^x \frac{1}{1+t^2} \, dt \\[9pt]  &= \int_0^x \left( 1 - t^2 + t^4 - t^6 + \cdots + (-1)^{n-1} t^{2n-2} + \frac{(-1)^n t^{2n}}{1+t^2} \right) \, dt \\[9pt]  &= \int_0^x \left( 1 - t^2 + t^4 - t^6 + \cdots + (-1)^{n-1} t^{2n-2} \right) \, dt + \int_0^x \frac{(-1)^n t^{2n}}{1+t^2} \, dt \\[9pt]  &= \sum_{k=0}^{n-1} \frac{(-1)^k x^{2k+1}}{2k+1} + \int_0^x \frac{(-1)^n t^{2n}}{1+t^2} \, dt \\[9pt]  &= \sum_{k=0}^{n-1} \frac{(-1)^k x^{2k+1}}{2k+1} + E_{2n}(x). \end{align*}

Therefore we have

    \[ E_{2n}(x) = \int_0^x \frac{(-1)^n t^{2n}}{1+t^2}. \]

So, we can bound the error term by bounding the integral,

    \begin{align*}  | E_{2n} (x) | &= \left| \int_0^x \frac{(-1)^n t^{2n}}{1+t^2} \, dt \right| \\[9pt]  &\leq \int_0^x \frac{t^{2n}}{1+t^2} \, dt \\[9pt]  &\leq \int_0^x t^{2n} \, dt &(1+t^2 \geq 1) \\[9pt]  &= \frac{x^{2n+1}}{2n+1}. \qquad \blacksquare \end{align*}

Prove an inequality for the error of the Taylor polynomial of sin x

Prove that the error of the Taylor expansion of \sin x satisfies the following inequality.

    \[ \sin x = \sum_{k=1}^n \frac{(-1)^{k-1} x^{2k-1}}{(2k-1)!} + E_{2n}(x), \qquad |E_{2n}(x)| \leq \frac{|x|^{2n+1}}{(2n+1)!}. \]


Proof. Since the derivatives of \sin x are always \cos x, -\sin x, -\cos x, or \sin x we know that for f(x) = \sin x we have |f^{(n+1)}(x)| \leq 1. (In other words, the n+1st derivative is bounded above by 1 and below by -1.) Therefore, we can apply Theorem 7.7 (p. 280 of Apostol) to estimate the error in Taylor’s formula at a= 0 with m = -1 and M = 1. For x > 0 this gives us

    \begin{alignat*}{3} & m \frac{(x-a)^{n+1}}{(n+1)!} &\leq \ \ E_n (x) &\leq M \frac{(x-a)^{n+1}}{(n+1)!} \\[9pt] \implies & -\frac{x^{n+1}}{(n+1)!} &\leq \ \ E_n (x) &\leq \frac{x^{n+1}}{(n+1)!} \\[9pt] \implies & -\frac{x^{2n+1}}{(2n+1)!} &\leq \ \ E_{2n} (x) &\leq \frac{x^{2n+1}}{(2n+1)!} \\[9pt] \implies & \phantom{-}|E_{2n}(x)| &\leq \frac{x^{2n+1}}{(2n+1)!}. \end{alignat*}

Next, (from the second part of Theorem 7.7) if x < 0 we have

    \begin{alignat*}{3}  & m \frac{(a-x)^{n+1}}{(n+1)!} &\leq (-1)^{n+1} E_n (x) &\leq M \frac{(a-x)^{n+1}}{(n+1)!} \\[9pt]  \implies & - \frac{(-x)^{n+1}}{(n+1)!} &\leq (-1)^{n+1} E_n (x) &\leq \frac{(-x)^{n+1}}{(n+1)!} \\[9pt]  \implies & - \frac{(-x)^{2n+1}}{(2n+1)!} &\leq (-1)^{2n+1} E_{2n}(x) &\leq \frac{(-x)^{2n+1}}{(2n+1)!} \\[9pt]  \implies & - \frac{(-x)^{2n+1}}{(2n+1)!} &\leq -E_{2n} (x) &\leq \frac{(-x)^{2n+1}}{(2n+1)!} \\[9pt]  \implies & \phantom{-} |E_{2n} (x)| &\leq \frac{x^{2n+1}}{(2n+1)!}. \qquad \blacksquare \end{alignat*}

Prove an inequality of exponentials

For all x, y > 0 and for any constants a,b such that 0 < a < b prove that

    \[ \big( x^b + y^b \big)^{\frac{1}{b}} <  \big( x^a + y^a \big)^{\frac{1}{a}}. \]


Proof. We want to consider the function

    \[ f(t) = (x^t + y^t)^{\frac{1}{t}}. \]

If we can show this function is decreasing on the positive real axis then we establish the inequality since this would mean that if 0 < a < b then

    \[ f(b) < f(a) \quad \implies \quad (x^b+y^b)^{\frac{1}{b}} < (x^a+y^a)^{\frac{1}{a}}.\]

(So, the trick here is to think of this as a function of the exponent. The x and y are some positive fixed constants.) To take the derivative of f(t) we use logarithmic differentiation,

    \begin{align*}  &&f(t) &= (x^t+y^t)^{\frac{1}{t}} \\[10pt]  \implies &&\log f(t) &= \frac{1}{t} \log (x^t+y^t) \\[10pt]  \implies &&\frac{d}{dt} (\log f(t)) &= \frac{d}{dt} \left( \frac{1}{t} \log (x^t+y^t) \right) \\[10pt]  \implies &&\frac{f'(t)}{f(t)} &= -\frac{1}{t^2} \log (x^t+y^t) + \frac{1}{t} \left( \frac{1}{x^t+y^t} \right) \left( x^t \log x + y^t \log y \right) \\[10pt]  \implies &&\frac{f'(t)}{f(t)} &= \frac{-\log(x^t+y^t)}{t^2} + \frac{x^t \log x + y^t \log y}{t (x^t+y^t)}. \end{align*}

Multiplying both sides by f(t) we then obtain

    \begin{align*}  f'(t) &= \left( \frac{-\log(x^t+y^t)}{t^2} + \frac{x^t \log x + y^t \log y}{t(x^t+y^t)}  \right) (x^t+y^t)^{\frac{1}{t}} \\[10pt]  &= (x^t + y^t)^{\frac{1}{t}} \left( \frac{t(x^t \log x + y^t \log y) - (x^t+y^t)\log(x^t+y^t)}{t^2 (x^t+y^t)} \right) \\[10pt]  &= \left( \frac{(x^t+y^t)^{\frac{1}{t} - 1}}{t^2} \right) \left( x^t t \log x + y^t t \log y - (x^t + y^t)\log(x^t+y^t) \right) \\[10pt]  &= \left( \frac{(x^t+y^t)^{\frac{1}{t}-1}}{t^2} \right) \left( x^t \log x^t + y^t \log y^t - x^t \log (x^t+y^t) - y^t \log (x^t+y^t) \right) \\[10pt]  &= \left( \frac{(x^t+y^t)^{\frac{1}{t} - 1}}{t^2} \right) \left( x^t \log \left(\frac{x^t}{x^t+y^t}\right) + y^t \log \left( \frac{y^t}{x^t+y^t} \right) \right). \end{align*}

Now we can conclude that f'(t) < 0 for all t > 0 since the first term in the product

    \[ \frac{(x^t+y^t)^{\frac{1}{t}-1}}{t^2} > 0. \]

Since x, y > 0 (any real power of a positive number is still positive) and t^2 > 0. For the second term we have

    \[ x^t \log \left( \frac{x^t}{x^t+y^t} \right) + y^t \log \left( \frac{y^t}{x^t+y^t} \right) < 0 \]

since x^t and y^t are positive, but both logarithms are negative. We know these logarithms are negative since

    \[ \frac{x^t}{x^t+y^t} < 1 \quad \text{and} \quad \frac{y^t}{x^t+y^t} < 1 \]

implies

    \[ \log \left( \frac{x^t}{x^t+y^t} \right) < 0 \quad \text{and} \quad \log \left( \frac{y^t}{x^t+y^t} \right) < 0. \]

Hence, f'(t) < 0 for all t > 0. This means f(t) is a decreasing function. Therefore, if 0 < a < b then we have

    \[ f(b) < f(a) \quad \implies \quad \big( x^b + y^b \big)^{\frac{1}{b}} < \big( x^a + y^a \big)^{\frac{1}{a}}. \qquad \blacksquare \]