8

So far, I have denoted $$\frac{MA}{MB}=a,\frac{BN}{NC}=b,\frac{CP}{PD}=c,\frac{DQ}{QA}=d,\frac{AR}{RC}=e,\frac{BS}{SD}=f$$ and then I have computed the volumes of the "small" tetrahedra $$V_1=V_{AMRQ}=\frac{ae}{(a+1)(e+1)(d+1)};$$ $$V_2=V_{BMSN}=\frac{bf}{(b+1)(f+1)(a+1)};$$ $$V_3=V_{CRNP}=\frac{c}{(c+1)(e+1)(b+1)};$$ $$V_4=V_{DPQS}=\frac{d}{(d+1)(c+1)(f+1)}.$$ Also, by Menelaus theorem for tetrahedra, one can deduce that $$abcd=1;~~~ af=ce; ~~~b=fde.$$ It remains to show that $$V_1+V_2+V_3+V_4\geq \frac{1}{2}.$$

The problem has a well known analogue in 2-D: for a triangle $ABC$ of area 1 and $AM, BN, CP$ concurrent cevians, the area of the triangle $MNP$ is less than 1/4.


Later edit/update:

I have managed to rewrite $V_k$ as follows: $$V_1={\frac {a}{ \left( a+1 \right) \left( d+1 \right) \left( cd+1 \right) }};$$ $$V_2={\frac {b}{ \left( a+1 \right) \left( b+1 \right) \left( ad+1 \right) }};$$ $$V_3= {\frac {c}{ \left( c+1 \right) \left( b+1 \right) \left( ab+1 \right) }};$$ $$V_4= {\frac {d}{ \left( c+1 \right) \left( d+1 \right) \left( bc+1 \right) }},$$ Plus, we still have $abcd=1$.

JohnnyC
  • 1,220
  • Good generalization, good question! I try to understand the symmetry beneath the 6 proportion you mentioned, I think the equations would become easier after this understanding.Furthermore I'm thinking a geometric way of proof. – MasM Feb 11 '21 at 21:49
  • Indeed, the problem seems intriguingly hard. It was on a short list of the National Olympiad in Romania for the 8th grade, a few years ago. – JohnnyC Feb 12 '21 at 10:32
  • 1
    Your question should be clear without the title. After the title has drawn someone's attention to the question by giving a good description, its purpose is done. The title is not the first sentence of your question, so make sure that the question body does not rely on specific information in the title. – Martin R Feb 13 '21 at 11:33
  • Pardon my ignorance: What does “concurrent” mean in this context? – Martin R Feb 13 '21 at 11:35
  • The lines $MP, NQ, RS$ have a common point. – JohnnyC Feb 13 '21 at 11:38
  • I will take into account in the future the point about the title of the question. Thanks! – JohnnyC Feb 13 '21 at 11:52
  • @JohnnyC It seems there should be an easier way to prove as you mentioned it is Olympiad question and even your reduced inequality to 4 variable seems hard enough to be an IMO P3 though I'm not expert in solving inequalities but check a 6×6 matrix with elements of 4multiplying variables of each$V_i$ and determinant of sum of $V_i$s then use the hadamard's inequality for determinant of a matrix. If it didn't workout you would propose this inequality in mathexchange again. – MasM Feb 14 '21 at 15:59
  • @JohnnyC I've just asked whether it is possible to use hadamard' inequality or not here: https://math.stackexchange.com/questions/4026020/determinant-is-given-how-to-find-a-matix-with-lowest-dimension-which-has-the-de?noredirect=1#comment8311877_4026020 – MasM Feb 15 '21 at 00:43
  • 2
    @MasM Thank you very much! Indeed, there has to be a simpler proof for the initial geometric result, if not for the inequality equivalent with it. Using the classic substitution $a=\frac{x}{y},b=\frac{y}{z},c=\frac{z}{t},d=\frac{t}{x}$ the inequality becomes even more nice & "symmetric": $\sum \frac{x^3}{(x+y)(x+z)(x+t)}\geq \frac{1}{2}$. – JohnnyC Feb 15 '21 at 11:07
  • @JohnnyC Well done, you solved it by your own, it would be great if you sum up your calculations and write down the solution. I searched for a completely geometrical solution but I failed to find, maybe because it needs more advanced theorems and tools beyond the knowledge of school Euclidean geometry. Thank for proposing this question. It has evoked some ideas for me personally that the question I've asked was one of them. – MasM Feb 15 '21 at 18:40

2 Answers2

4

Let $X$ be the intersection of the three segments $MP$, $NQ$, $RS$. Since $ABCD$ is non-degenerate, its volume is one, the points $A,B,C,D$ are not in a plane, so we can uniquely write $$ X = aA+bB+cC+dD\ , \qquad a+b+c+d=1\ . $$ math stackexchange 4016106 Then grouping pairs of vertices, we obtain formulas for $M,N,P,Q,R,S$ in terms of these weights, corresponding to the groupings: $$ \begin{aligned} X &= (a+b)\cdot \underbrace{\frac {aA+bB}{a+b}}_{M} + (c+d)\cdot \underbrace{\frac {cC+dD}{c+d}}_{P} \\[2mm] &= (a+c)\cdot \underbrace{\frac {aA+cC}{a+c}}_{R} + (b+d)\cdot \underbrace{\frac {bB+dD}{b+d}}_{S} \\[2mm] &= (a+d)\cdot \underbrace{\frac {aA+dD}{a+d}}_{Q} + (b+c)\cdot \underbrace{\frac {bB+cC}{b+c}}_{N} \ . \end{aligned} $$ (They really know in Romania vectors in 3D in the VIII.th class. The relations the children could have been written involve the vectors $OX$, $OA$, $OB$, $OCV$, $OD$ instead of the above.)

So $M$ is situated on $AB$ such that $M=\frac 1{a+b}(aA+bB)$, which gives $AM:AB=b:(a+b)$. We write similar relations for $AR:AC=c:(a+c)$ and $AQ:AD=d:(a+d)$. So the volume $[AMRQ]$, seen as a fraction of the volume $[ABCD]=1$ is $$ [AMRQ] = \frac{AM}{AB}\cdot \frac{AR}{AC}\cdot \frac{AQ}{AD}\cdot [ABCD] =\frac{bcd}{(a+b)(a+c)(a+d)}\cdot 1\ . $$ So the given geometric inequality is equivalent to the algebraic inequality $$ \sum\frac{bcd}{(a+b)(a+c)(a+d)} \ge \frac 12\ . $$ The sum has four terms obtained by acting with the powers of the cyclic permutation $(a,b,c,d)$ on the written term.


I could find quickly only the brute force proof. We multiply with the common denominator and have to show the domination: $$ \underbrace{\sum a^3b^2c}_{4\cdot 3\cdot 2=24\text{ terms}} + 2\underbrace{\sum a^2b^2c^2}_{4\text{ terms}} \ge 2\underbrace{\sum a^3bcd}_{4\text{ terms}} + 4\underbrace{\sum a^2b^2cd}_{6\text{ terms}} \ . $$ The domination is clear, the monominal type $(3,2,1,0)$ dominates $(3,1,1,1)$ and $(2,2,1,1)$. And $(2,2,2,0)$ dominates $(2,2,1,1)$. So one can stop here.

But for the convenience of the reader, let us write this explicitly. First of all $$ b^2c+c^2b+b^2d+d^2b + c^2d+d^2c\ge 6bcd\ . $$ (Arithmetic-geometric mean.) We multiply with $a^3$ and sum. It remains (i.e. is sufficient) to show $$ \frac 23 \underbrace{\sum a^3b^2c}_{4\cdot 3\cdot 2=24\text{ terms}} + 2\underbrace{\sum a^2b^2c^2}_{4\text{ terms}} \ge 4\underbrace{\sum a^2b^2cd}_{6\text{ terms}} \ . $$ Now use $a^3b^2c+ab^2c^3\ge 2a^2b^2c^2$ to estimate downwards the LHS with a scalar multiple of $\sum a^2b^2c^2$, to have a simpler LHS, so $$ \begin{aligned} \frac 23 \underbrace{\sum a^3b^2c}_{4\cdot 3\cdot 2=24\text{ terms}} + 2\underbrace{\sum a^2b^2c^2}_{4\text{ terms}} &\ge \frac 23\cdot 6 \underbrace{\sum a^2b^2c^2}_{4\text{ terms}} + 2\underbrace{\sum a^2b^2c^2}_{4\text{ terms}} = 6 \underbrace{\sum a^2b^2c^2}_{4\text{ terms}} \\ &\ge 4\underbrace{\sum a^2b^2cd}_{6\text{ terms}} \ . \end{aligned} $$ The last inequality follows from $a^2b^2c^2+a^2b^2d^2\ge 2a^2b^2cd$.

$\square$


Note: The problem was proposed by Flavian Georgescu, Bucharest, at that time college student (elev), so maximally 18yo, and it should be doable with the aid of "thin air" available in the 8.th class (14yo students) of the Romanian mathematical matter from the school book. I suppose Titu or a mean inequality was the solution of the author, if he did it as above, else i see only the possibility of a geometrical argument using $[XMRQ]$ and the similar tetraedra.

dan_fulea
  • 32,856
  • Nice +1. I am a bit lost with Muirhead and the domination though. What does it mean ( 24 terms, 6 terms,... ) ? Are those sums symmetric ? If yes the coefficients do not seem correct to apply Muirhead. Probably the sums are not symmetric ? I understand that 24 = 6×4 but do not get what the notation means... – Thomas Jul 22 '21 at 09:12
  • @Thomas I hat to reread my computations... In $\sum a^3b^2c$ we have all terms of the shape $x^3y^2z$ with coefficients equal to one, where the the "letters" $x,y,z$ are different and among $a,b,c,d$. (So for the choice of the $x$-letter we have $4$ possibilities, for a fixed $x$ we have for $y$ the remained $3$ possibilities in each case, and for fixed $(x,y)$ there are two possibilities for $z$.) In the sum $\sum a^2b^2c^2$ we have four terms $x^2y^2z^2$, we fix the missing letter and use the other three squared, so explicitly $\sum a^2b^2c^2=a^2b^2c^2+a^2b^2d^2+a^2c^2d^2+b^2c^2d^2$. – dan_fulea Jul 23 '21 at 13:13
  • @Thomas Then similarly $\sum a^3bcd=abcd\sum a^2=abcd(a^2+b^2+c^2+d^2)$, there are $4=\binom 41$ chances for the choice of the letter cubed. And for the last sum, $\sum a^2b^2cd=abcd\sum ab=abcd(ab+ac+ad+bc+bd+cd)$ there are six terms for the $\binom 42$ possibilities to pick the letters to be squared in $\sum a^2b^2cd$. – dan_fulea Jul 23 '21 at 13:18
  • Ah ok so your sum notation considers the sum of the monomials that you would have in a standard symmetric sum, with prefactor one and not considering the "degeneracy" due to all permutations that leads to coefficients larger than one if all symmetric terms are considered – Thomas Jul 23 '21 at 20:00
  • In this sense, it is not the standard symmetric notation that you have in Muirhead. How can you drive than conclusions about the domination without further computations? It looks you have some quick method to pass from your notation to the symmetric one? – Thomas Jul 23 '21 at 20:02
1

Well this not a rigorous proof for this problem, but I decided to express some of my ideas that may lead to a geometric proof.

First I managed to find set of points of concurrency $S_i$ inside an arbitrarily triangle $T$ which make same area of triangles shaped by their three concurrent segments as you mentioned in your 2-D case. the sets of such points can also be defined by the level sets of a function $f$ that maps each point inside of the triangle $T$ to the area of its mentioned triangle. by use of continuity we can deduce that every level set is a jordan closed loop $J_i,i\in [0,1]$ and even it seems that they are convex loops which doesn't intersect eachother and their value by function $f$ (which resembles the mentioned triangles area) become bigger and bigger as the loops become smaller and smaller until they retract to centroid of triangle $T$.

I think this process can be easily generalized to the case of the the 3-d tetrahedron and the level sets here would become border of convex bodies which retracts to the centroid of tetrahedron when it gains its highest value by the mapping function sending each point inside of tetrahedron to the volume of its octahedron you mentioned.

While searching for a proof of this claim ,I was curious about a transformation that convert an arbitrarily tetrahedron to a regular one as it seems good idea too for the 2-D case to convert a given triangle to the most symmetrical triangle which is equilateral triangle. I've just find a kind of such transformation here. I still don't know this transformation would be useful from the perspective of my approach for the sake of the statement of inequality of areas in this problem. for example if the jordan loops $J_i$ in $T$ becomes circles while transformed, then the proof would become easier.

My second idea is about to prove this statement:

The volume of mentioned octahedron $MNPQRS$ will be maximum if and only if the centroids of $ABCD$ and $MNPQRS$ coincide.

MasM
  • 641
  • It is also interesting to see many inequalities like the one here is related to the transformation I mentioned here and it was known an formulated in the link: https://math.stackexchange.com/a/542684/346943 ses 18 question related to the answer! – MasM Feb 18 '21 at 14:44