I ntrnat. J. Mah. Math. Sci. Vol. 5 No. 2 (1982) 337-343 337 THE POWER MEAN AND THE LOGARITHMIC MEAN CHRISTOPHER OLUTUNDE IMORU Department of Mathematics University of Ife Ile-Ife Oyo State NIGERIA (Received July 3, 1979 and in revised form, March 24, 1980) ABSTRACT. In a very interesting and recent note, Tung-Po Lin [I] obtained the least value q and the greatest value p such that M P <L<M q is valid for all distinct positive numbers x and y where Ms (x +.y.)s 2 and L In x-ln Y The object of this paper is to give a simpler proof than Lin’s of a more general result. and ha, More precisely, the author obtained the classes of functions f agR such that In f (t) KEY WORDS AND PHRASES. functions. h a (t) [tl/ + I] - > O, t > I. Concave Functions, Space of Lebesque -integrable 1980 MATHEMATICS SUBJECT CLASSIFICATION CODES. Primary 26 A 86. 1. INTRODUCTION. There is an extensive literature on the extensions and generalizations of inequalities involving the power mean and the logarithmic mean and their applica- C.O. IMORU 338 [see 2-5]. tions Tung-Po Lin [I] considered the least value p and the greatest value q such that M q <L<M (1.1) P is valid for all distinct positive real numbers x and y, where L x-y L(x,y) Inx-lny and H M.(x,y) a (X +2 "y c e R are respectively the logarithmic mean and the power mean of order of two dis- tinct positive numbers x and y. Indeed, Lin showed that q are the best possible values one can 0 and p get in order that (I.I) may be satisfied for any distinct positive real numbers x and y. This result is sharper than an earlier inequality due to Carlson [6], namely: M O < L < for any distinct x, y > o M (I 2) As observed by Lin (If], Theroems 3 and 4), if 0<p<I/3, there exist x > 0 and y > 0 such that L(x,y) > M (x,y). On the other hand, there exist x > 0 and P y > 0 such that L(x,y) < M (x,y). In fact, if we suppose that x > y > o, which P is perfectly in order due to the symmetry of L(x,y) and M (x,y) in x and y, one P can readily show that L(x,y) < Mp(x,y), o < p < - . (1.3) For is large enough. The reversed inequality holds for smaller values of Y Y instance, numerical calculations with the aid of a computer easily show that the x result (1.3) holds for p if is greater than (9 9) 4 (25 3799) s I0 Y if ,, (56-6188) 6 (119-7628) 7 (246.8528) 8 (501.9099) 9 and (1012.9460) 1 respectively. . From these results, we can readily compute the set {(x,y); x > o and y > o} for which H set {(x,y); P (x,y) < L(x,y) X < > p > (I012.9460)I}. For example, M It is interesting to ne (x,y) < L(x,y) on the that the above esti- POWER MEAN AND LOGARITHMIC MEAN x mates show that 339 o, which is compatible with the left-hand side of in- as p Y equality (I.I), namely: M o (x,y) < L(x,y) for distinct x > y > o. I/p more than doubles from p Furthermore, the critical value of (x/y) p I/(5+n), where (x/y)I/P values of o, I, 2, n 5. 1/(4+n) to It is our conjecture that the critical is more than double from p I/n to I/(n+l), n=l, n, The purpose of this note is to give a simpler proof than Carlson or Lin of a Indeed, we obtain in Section 2 the classes of functions f more general result. and h e R, for which the function ff F (t)= In f (t) -ha (t) (t I/ + I) t > This leads to the determination of a polynomial of degree is strictly positive. n, (n.>l) which is - a lower bound of the function ()n In t. The compari- with the ones given by Carlson and Lin show I, 2, son of these bounds for n H(t) that not only are our results more general in form, but they are somewhat sharper. We point out that Carlson’s inequalities (1.2) are special cases of an inequality for convex and concave functions. Indeed, from the Jensen inequality for concave functions, t 1 f(x) d(x) t fl where dp(x) R, ff(x)} t fl dp(x) (1.4) d(x) L(R, d), the space of Lebesque p is concave and fg tions on O- > we obtain, on setting f(x) , p(x) x and integrable func- (x) x I/p p > or p < o, the inequality In t t-1 Since o < p < I. (1.4) > is reversed if Taking p 2 in tl-1/P-1]p Z [_p__ p-1 is (1.5), P > orp < o. (1.5) convex, inequality (1.5) is reversed if we obtain the right-hand side of inequality (1.2). 2. MAIN RESULTS. We pointed out that Lin, with the aid of the Maclaurin series, showed that 340 C.O. IMORU the best possible value of p for which L < M A proof of this interesting " is p Ill result without the Maclaurin series was indicated in the paper Following Flander’s approach, we obtain the following result. Flanders. F(z) Let Theorem 2.1: in ha.(1) (z+l) (z+l) o is differentiable in R. R, where Suppose fo is an f(1) Then F (z) > 0, for every z > I, provided 0. o +I hl (z) f (z) f Proof: ha(z) f(z) ho increasing positive function and and - by Harley + (z+l) a h(z) f(z) (z) > o. (2.1) (z) f Computing the first derivative of f (z) yields Fl(z) Ac(z) (z+l) (+1 Gcx(z)] (2 .2) where f (z) n (z) f (z) (z+l) a+l and hcl(z) Ga(z) (z+l) -a h((z) f(x(z) (z) f (z) f 0 if and only if (z+l) a+l Observe that F (z) Fo(z) fo(z) G Fa(1) 0 and z > I. This completes the proof of the theorem. O(z) > 0, z I. Since > 0 wherever z > I, we conclude that F (z) > 0, for all Corollary 2.2: - For every a E Inz k=l is valid provided z c+l a + k R, 8k (zk-1) 5 k=2 + [{+1 X+lCk +--a81 + < k < < z+l) k-1 -k-1 k k=l , a, where a z > 1, [k-l+k ]]z --]z+l +, the inequality (2.3) k a > 0 (2.4) POWER MEAN AND LOGARITHMIC MEAN a 5k=l Let h (z) Proof" result follows at once. 13k (z k_ I) and f (z) z a 341 in Theorem 2 - - - Then the Observe that for all z > I, 5k=l satisfies the conditions 13k 130 --> 2 z k_ 1) O n n z 13 k u-1 5k=l and The inequality (2.5) is strict unless quently, we have => 2c (za-1), zk 1) and k => (13k)l<=k=< (2.5) (2.6) o. o, k z-1) < (z+l )a 5- z u 2 13a provided < k < u. Conse- a.[ k=l (2.7) wherever (2.4) and (2.6) are valid. Hence for all z > I, the polynomial Pa(z) {/ (13k)l<k<ot (2.6), satisfies :Y y Inx-lny < k=l 13k (xk/a yk/U) k 2(In < x (z k- 1 k=l where z+l- c (--) H(z) In z, K(z) Suppose condition (2.6) holds. x a is a lower bound of the function as well as an upper bound of the function Corollary 2.3: 2 For x > y > 0, the inequality ,x I/a + 2 In y) yl/x) e’, (2.8) is valid whenever +I + x 5k=2 [u + [u +I- Proof: +1 f l31 Ck + f S_ k=l Bk-1 k- l+k] xk/ay(a+l-k)/a a k-1 _._.]=lk x 1/ay (o+ The result is immediate on setting z inequa ity (2.7). )/c > o x _q y where x > y > o in (2.9) 342 C.O. IMORU The cases a 2 and a 3 with a 2 k k and son’s result and Lin’s inequality respectively. o Suppose k # a lead to Carl- a a2a(1+/2)’ then inequality (2.8) gives an improved lower bound for the power mean ject to the condition (2.9). > o; H1/a, sub- To see this, let us consider the following special cases. Suppose a 1. Then inequality (2.8) yields x Inx- Iny + + 4(&2+2)]; provided x > y [1 Suppose a Inx Iny ’i’nx Iny + y < x > o < (1 +) z [1++4(2+2)]; (2.11) > o. 3, then inequality (2.8) yields If we finally consider the case when 3 x y Inx- Iny < (1 + We pointed out in Section > (9.9) 4 )2 y x Inx Iny < (x 3 3 + y (2.12) 2 ) + 4(z+2) + 412(z+2)+2(I+)4(+2)]; provided x > y(l+ Y (2 10) 2 2, we have Inx- Iny wherever x > y y x < (1 + > o. that inequality (1.3) holds for p % if It is pertinent to add here that these values can be obtained by 4. To illustrate the usefulness in the inequality (2.9) when Y of this method, we compute the set on which inequality (1.3), with p %, is true. solving for Take 2 k 4 in inequality (2.8). and o, k x y Inx- Iny provided z s 11 z 4 + 10 z 3 + 10 z < (x 11 z + + (2.13) 2 > o, where z 4 hand side of this inequality has the factorization (z + 1) (z-l) (z-5+246)(z-5-246). The result is x/y. The left POWER MEAN AND LOGARITHMIC MEAN 343 Consequently, inequality (2.13) is valid provided either > ()x L(x,y) < for a > 5 256 Mk(x,y) 0.I0 or on the set (_Zyx)k > 5 + [(x,y); 256 9.9. > (9.9)4]. Since x > y > o, we have Similar results can be obtained 5, 6, 7, Further specializations of the constants 8k’ < k < a, may be of interest in the determination of improved upper and lower bounds for the logarithmic mean and the power mean respectively. REFERENCES I. Tung-Po Lin: 2. E.B. Leach and M.C. Sholander: The Power Mean and the Logarithmic Mean; Amer. Math. Monthly 81 (1974), 879-883. Extended Mean Values; Amer. Math. Monthly 85 (1978), 84-90. 3. E.B. Leach and M.C. Sholander: Corrections to "Extended Mean Values"; Amer. Math. Monthly 85 (1978), 656. 4. K.B. Stolarsky: Generalizations of the Logarithmic Mean; Amer. Math. Mag. 48 (1975), 87-92. 5. K.B. Stolarsky: Monthly, 6. B.C. Carlson: The Power and Generalized Logarithmic to appear. The Logarithmic Mean; Amer. Math. Mean; Amer. Math. Monthly 79 (1972), 615-618.