Unofficial archive — problems, solutions & results © IMC, reproduced with permission.

IMC / 2003 / Problems / Day 1, P6

IMC 2003 · Day 1 · P6

very hard

Let f(z)=anzn+an1zn1++a1z+a0f(z) = a_n z^n + a_{n-1} z^{n-1} + \dots + a_1 z + a_0 be a polynomial with real coefficients. Prove that if all roots of ff lie in the left half-plane {zC:Rez<0}\{ z \in \mathbb{C} : \operatorname{Re} z < 0 \} then akak+3<ak+1ak+2a_k a_{k+3} < a_{k+1} a_{k+2} holds for every k=0,1,,n3k = 0, 1, \dots, n-3.

Solution (official)

The polynomial ff is a product of linear and quadratic factors, f(z)=i(kiz+li)j(pjz2+qjz+rj)f(z) = \prod_i (k_i z + l_i) \cdot \prod_j (p_j z^2 + q_j z + r_j), with ki,li,pj,qj,rjRk_i, l_i, p_j, q_j, r_j \in \mathbb{R}. Since all roots are in the left half-plane, for each ii, kik_i and lil_i are of the same sign, and for each jj, pj,qj,rjp_j, q_j, r_j are of the same sign, too. Hence, multiplying ff by 1-1 if necessary, the roots of ff don't change and ff becomes the polynomial with all positive coefficients.

For the simplicity, we extend the sequence of coefficients by an+1=an+2==0a_{n+1} = a_{n+2} = \dots = 0 and a1=a2==0a_{-1} = a_{-2} = \dots = 0 and prove the same statement for 1kn2-1 \le k \le n-2 by induction.

For n2n \le 2 the statement is obvious: ak+1a_{k+1} and ak+2a_{k+2} are positive and at least one of ak1a_{k-1} and ak+3a_{k+3} is 0; hence,

ak+1ak+2>akak+3=0a_{k+1} a_{k+2} > a_k a_{k+3} = 0.

Now assume that n3n \ge 3 and the statement is true for all smaller values of nn. Take a divisor of f(z)f(z) which has the form z2+pz+qz^2 + pz + q where pp and qq are positive real numbers. (Such a divisor can be obtained from a conjugate pair of roots or two real roots.) Then we can write f(z)=(z2+pz+q)(bn2zn2++b1z+b0)=(z2+pz+q)g(x).(1)\tag{1} f(z) = (z^2 + pz + q)(b_{n-2} z^{n-2} + \dots + b_1 z + b_0) = (z^2 + pz + q) g(x). The roots polynomial g(z)g(z) are in the left half-plane, so we have bk+1bk+2<bkbk+3b_{k+1} b_{k+2} < b_k b_{k+3}

for all 1kn4-1 \le k \le n-4. Defining bn1=bn==0b_{n-1} = b_n = \dots = 0 and b1=b2==0b_{-1} = b_{-2} = \dots = 0 as well, we also have bk+1bk+2bkbk+3b_{k+1} b_{k+2} \le b_k b_{k+3}

for all integer kk.

Now we prove ak+1ak+2>akak+3a_{k+1} a_{k+2} > a_k a_{k+3}. If k=1k = -1 or k=n2k = n-2 then this is obvious since ak+1ak+2a_{k+1} a_{k+2} is positive and akak+3=0a_k a_{k+3} = 0. Thus, assume 0kn30 \le k \le n-3. By an easy computation, ak+1ak+2akak+3==(qbk+1+pbk+bk1)(qbk+2+pbk+1+bk)(qbk+pbk1+bk2)(qbk+3+pbk+2+bk+1)==(bk1bkbk2bk+1)+p(bk2bk2bk+2)+q(bk1bk+2bk2bk+3)++p2(bkbk+1bk1bk+2)+q2(bk+1bk+2bkbk+3)+pq(bk+12bk1bk+3).\begin{align*} a_{k+1} &a_{k+2} - a_k a_{k+3} = \\ &= (q b_{k+1} + p b_k + b_{k-1})(q b_{k+2} + p b_{k+1} + b_k) - (q b_k + p b_{k-1} + b_{k-2})(q b_{k+3} + p b_{k+2} + b_{k+1}) = \\ &= (b_{k-1} b_k - b_{k-2} b_{k+1}) + p (b_k^2 - b_{k-2} b_{k+2}) + q (b_{k-1} b_{k+2} - b_{k-2} b_{k+3}) + \\ &\quad + p^2 (b_k b_{k+1} - b_{k-1} b_{k+2}) + q^2 (b_{k+1} b_{k+2} - b_k b_{k+3}) + pq (b_{k+1}^2 - b_{k-1} b_{k+3}). \end{align*} We prove that all the six terms are non-negative and at least one is positive. Term p2(bkbk+1bk1bk+2)p^2 (b_k b_{k+1} - b_{k-1} b_{k+2}) is positive since 0kn30 \le k \le n-3. Also terms bk1bkbk2bk+1b_{k-1} b_k - b_{k-2} b_{k+1} and q2(bk+1bk+2bkbk+3)q^2 (b_{k+1} b_{k+2} - b_k b_{k+3}) are non-negative by the induction hypothesis.

To check the sign of p(bk2bk2bk+2)p (b_k^2 - b_{k-2} b_{k+2}) consider bk1(bk2bk2bk+2)=bk2(bkbk+1bk1bk+2)+bk(bk1bkbk2bk+1)0.b_{k-1} (b_k^2 - b_{k-2} b_{k+2}) = b_{k-2} (b_k b_{k+1} - b_{k-1} b_{k+2}) + b_k (b_{k-1} b_k - b_{k-2} b_{k+1}) \ge 0. If bk1>0b_{k-1} > 0 we can divide by it to obtain bk2bk2bk+20b_k^2 - b_{k-2} b_{k+2} \ge 0. Otherwise, if bk1=0b_{k-1} = 0, either bk2=0b_{k-2} = 0 or bk+2=0b_{k+2} = 0 and thus bk2bk2bk+2=bk20b_k^2 - b_{k-2} b_{k+2} = b_k^2 \ge 0. Therefore, p(bk2bk2bk+2)0p (b_k^2 - b_{k-2} b_{k+2}) \ge 0 for all kk. Similarly, pq(bk+12bk1bk+3)0pq (b_{k+1}^2 - b_{k-1} b_{k+3}) \ge 0.

The sign of q(bk1bk+2bk2bk+3)q (b_{k-1} b_{k+2} - b_{k-2} b_{k+3}) can be checked in a similar way. Consider bk+1(bk1bk+2bk2bk+3)=bk1(bk+1bk+2bkbk+3)+bk+3(bk1bkbk2bk+1)0.b_{k+1} (b_{k-1} b_{k+2} - b_{k-2} b_{k+3}) = b_{k-1} (b_{k+1} b_{k+2} - b_k b_{k+3}) + b_{k+3} (b_{k-1} b_k - b_{k-2} b_{k+1}) \ge 0. If bk+1>0b_{k+1} > 0, we can divide by it. Otherwise either bk2=0b_{k-2} = 0 or bk+3=0b_{k+3} = 0. In all cases, we obtain bk1bk+2bk2bk+30b_{k-1} b_{k+2} - b_{k-2} b_{k+3} \ge 0.

Now the signs of all terms are checked and the proof is complete.

How the field did

contestants scored
185
average (of 20)
3.10
solved (≥ 80%)
8.1%
near-0 (≤ 10%)
74.1%
discrimination
0.44

Score distribution (field cohort)

Computed on contestants with a meaningful total (field cohort); discrimination is the corrected item–total correlation.

Similar problems

IMC 2020 · Day 1 · P4very hardavg 0.7/10 · solved 6% · near-0 92% · disc 0.34
IMC 2015 · Day 2 · P10killeravg 0.1/10 · solved 0% · near-0 98% · disc 0.26