Peter Y. Woo, 8/20/2016
Biola University, La Mirada, Ca. 90639
Problem 4064. (Proposed by Michel Bataille)
Let Z be a circle with center O. Let ABC be a triangle whose sides do not cross over O. Let A',B',C' be the poles of the lines BC, CA, AB relative to Z.
Prove that OA'.B'C'/(OA/BC) = OB'.C'A'/(OB.CA) = OC'.A'B'/(OC.AB). . . (i)
Proof. Let OD, OE, OF be ^s from O to BC, CA, AB respectively. Let OD, OE, OF have lengths a,b,c.
Let OA, OB, OC divide the angles of DABC into 6 angles, named cyclically, a,b,g,d,e,z as shown in the diagram.
Then the polar of line AB is C' which lies on OE, and similarly for the polar B' of line CA and the polar A' of line BC.
Consequently A is the pole of B'C', B is the pole of C'A', and C is the pole of A'B'.
Now OEAF is a cyclic quad, with OA as diameter of its circumcircle. Hence the angles E and F inside DOEF are a and b also.
WLOG, assume circle Z has radius 1.Then OB' = 1/b, OC' = 1/c
\ DOEF is similar to DOB'C'.
Now OE/OC' = c2, OF/OB' = b2,
\ the sizes of the two simiar triangles have radio bc : 1.
Hence B'C' = (1/bc) EF = (1/bc) b Sin a+b)/Sin a
Hence OA' . B'C' /(OA . BC)
= (1/abc)b Sin (a+b)/Sin a / [(b/Sin a) BC]
= (1/abc) Sin (a+b) / BC = 1/(2R abc) . . . (ii)
where R is the circumradius of DABC.
Similarly, we can prove the other two expressions of (i) are all = (ii), due to symmetry.
Hence (i) is proved.(QED)
Peter Y. Woo, 3/04/2017
Biola University, La Mirada, Ca. 90639
Problem 4129. (Proposed by Lorean Saccanu).
Let ABC be an acute-angled triangle and let k = 3(2-Ö3). Prove that
sec A + sec B + sec C ³ k + tan A + tan B + tan C.
Proof. Let a=A/2, b=B/2, g=C/2.
We shall prove f(a) + f(b) + f(c) ³ k . . . (i)
where f(f) =def sec 2f - tan 2f. Let s = sinf, c = cosf, t = tanf.
Then f(f) = (1- sin2f)/cos2f (using trig formulas like crazy)
= [(s2+c2) - 2 s c ] / (c2 -s2) = (c-s)/(c+s)
= 1/[ (1+t)/(1-t)] = cot (45°+f). (Most USA highschoolers do not know this goodie formula)
Now a, b, g are between 0 and p/2, their average being p/6 and sum being p/2.
Hence on the xy-plane the points P(a, f(a)), P'(b, f(b)), P"(g, f(g)) lie on the curve
y = cot(45°+x) that is concave upwards.
Hence the geometric centroid of triangle PP'P" is (p/3, y'),
where y' is (1/3)(cot(45° +a) +cot(45°+b) +cot(45°+g))
and must be ³ cot(45°+p/6) = sec(30°) - tan(30°) = k.
The whole problem is thus proved.
Peter Y. Woo, 4/02/2017
Biola University, La Mirada, Ca. 90639
Problem 4197. (Proposed by Michel Bataille)
Let x,y,z all > 0 satisfying xy +yz +zx +2xyz = 1. . .(i)
Prove that (a) 1/x +1/y +1/z ³6 and (b) x+y+z ³ 3/2.
Proof. Let k3 = xyz.
Then xy +yz +zx £ 3k2, so that
3 k2 + 2 k3 £ 1.
Hence k £ 1/2.
Now from (i) we get 1/x +1/y +1/z + 2 = 1/k3 which³ 8.
Hence (a) is proved.
Again, from (i) xy + yz +zx = 1 -2k3 which ³ 1 -1/4 = 3/4.
Hence (x+y+z)2 = (x2+y2+z2)+2(xy+yz+zx) ³ 3(xy+yz+yz) which ³9/4.
Hence x+y+z ³ Ö(9/4) = 3/2, thus proving (b). (QED)
Peter Y. Woo, 1/10/2018
Biola University, La Mirada, Ca. 90639
Problem 4240. (Proposed by Rozenberg & Giugiuc.)
Let a,b,c be all > 0 with a+b+c = 1/a +1/b +1/c . . . (i)
Prove that 1+a+b+c ³ 4abc . . . . (ii)
Proof. WLOG, assume a < b < c. We shall write a+b/c+d to mean a + (b/c) + d because the symbols / and * have stronger binding power than + and -.
It is impossible for a,b,c to be all > 1 or all < 1. Otherwise (i) cannot be satisfied. So we have only 2 cases: Case 1: a £ b £ 1 < c or Case 2: a < 1 £ b £ c .
Strategy: Observe 1/a+1/b +1/c = (ab + bc + ca)/abc,
hence 4abc = 4(ab+bc+ca)/(1/a + 1/b + 1/c) = 4(ab+bc+ca)/(a+b+c) . . . (iii)
On the other hand, 1 + a + b + c = (a+b+c+a2+b2+c2+2ab+2bc+2ca) / (a+b+c) . . (iv)
To prove (iii) < (iv) is to prove
a+b+c+a2+b2+c2 - 2(ab+bc+ca) > 0 . . . (v)
There is a function we need to familiarise: y = x - 1/x = f(x) for 0 < x < ¥. It is one arm of a hyperbola, with asymtotes being
y = x and y = 0. So it is a convex curve for 0 < x < ¥, with f(1) = 0.
The requirement a + b + c = 1/a + 1/b + 1/c is equiv. to f(a) + f(b) + f(c) = 0.
In particular case, if a < b < 1 < c, then we can
replace b by 1 and c by d = 1/a such that f(b) + f(c) = f(1) + f(d),
and because of convexity of the curve y = f(x), d has to be less than c,
namely d satisfies d = 1/a so that f(d) = - f(a) and f(1) = 0,
thus satistying f(a) + f(b) + f(c) = f(a) + f(1) + f(1/a) ,
and 1 + d < b + c due to convexity. This trick shall be called Lemma 1.
Now this is case 1: LHS(v) = a + b + c + a2 +(b-c)2-2a(b+c) using Lemma1,
> a + (1 + 1/a) + a2 + (b-c)2 -2a (1+1/a)
= (b-c)2 + (1/a) [a2 + a + 1 + a3 -2a2 -2a]
= (b-c)2 + (1/a) (1-a) (1-a2) > 0, thus (v) is true.
Again for case 2: a < 1 < b < c, replace a by 1/c and b by 1
then LHS(v) = (a + b) + c + c2 +(a-b)2 -2c(a+b), using Lemma 1,
> (1+1/c) + c + c2 -2c(1+1/c) + (a-b)2
= (a-b)2 + (1/c)[ c - 1 - c2 + c3 ]
= (a-b)2 + (1/c) (c-1) (1 +c2 ) > 0, thus (v) is true.
We have proved (v). QED
Peter Y. Woo, 6/15/2017
Biola University, La Mirada, Ca. 90639
Problem. Let a+b+c+d = 1/a +1/b +1/c +1/d . . . (i)
Find the lower bound of abc+ abd+ acd + bcd.
Proof/. Let a =b =c = e, a very small number compared with 1.
Then 1/a +1/b +1/c = 3/e, = d- 1/d + 3e.
Hence d < and very close to 3/a = 3/e.
Hence abc + abd +acd +bcd = (1/a + 1/b +1/c + 1/d) (abcd)
= (a+b+c+d) (abcd) < (3e + 3/e) e3 3/e = 9 (e3 + e)
which is arbitrarily small.
Hence the lower bound is zero.