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

IMC / 2025 / Problems / Day 1, P3

IMC 2025 · Day 1 · P3

easy

Denote by SS the set of all real symmetric 2025×20252025 \times 2025 matrices of rank 1 whose entries take values 1-1 or +1+1. Let A,BSA, B \in S be matrices chosen independently uniformly at random. Find the probability that AA and BB commute, i.e. AB=BAAB = BA.

(proposed by Marian Panţiruc, ”Gheorghe Asachi” Technical University of Iaşi, Romania)

Solution (official)

Let n=2025n = 2025. First, we give a charaterisation of matrices in SS.

Suppose that A=(aij)i,j=1nSA = (a_{ij})_{i,j=1}^{n} \in S. Since rkA=1\operatorname{rk} A = 1, for every 1<i,jn1 < i, j \le n, we have det(a11a1jai1aij)=a11aijai1a1j=a11aijai1aj1=0.\det \begin{pmatrix} a_{11} & a_{1j} \\ a_{i1} & a_{ij} \end{pmatrix} = a_{11} a_{ij} - a_{i1} a_{1j} = a_{11} a_{ij} - a_{i1} a_{j1} = 0. If a11=1a_{11} = 1 then this means aij=ai1aj1a_{ij} = a_{i1} a_{j1}. In this case, let u=(a11a21an1)=(1a21an1)u = \begin{pmatrix} a_{11} \\ a_{21} \\ \vdots \\ a_{n1} \end{pmatrix} = \begin{pmatrix} 1 \\ a_{21} \\ \vdots \\ a_{n1} \end{pmatrix}; then we have A=(ai1aj1)=uuA = \bigl( a_{i1} a_{j1} \bigr) = u u^\top. Otherwise, if a11=1a_{11} = -1, we have aij=ai1aj1a_{ij} = -a_{i1} a_{j1}. In that case, let u=(a11a21an1)=(1a21an1)u = \begin{pmatrix} -a_{11} \\ -a_{21} \\ \vdots \\ -a_{n1} \end{pmatrix} = \begin{pmatrix} 1 \\ -a_{21} \\ \vdots \\ -a_{n1} \end{pmatrix}; then A=(ai1aj1)=uuA = -\bigl( a_{i1} a_{j1} \bigr) = -u u^\top.

Hence, all matrices in SS can uniquely be written as ±uu\pm u u^\top with a vector u=(1u2un)u = \begin{pmatrix} 1 \\ u_2 \\ \vdots \\ u_n \end{pmatrix} such that u2,,un{±1}u_2, \dots, u_n \in \{\pm 1\}. (Note that rk(uu)=1\operatorname{rk}(u u^\top) = 1 is satisfied.) In particular, we have S=2n|S| = 2^n, because the sign and the coordinates u2,,unu_2, \dots, u_n can be chosen independently.

Now, if A=±uuTA = \pm u u^T and B=±vvTB = \pm v v^T are elements of SS, then AB=±(uu)(vv)=±u(uv)v=±uu,vv=±u,v(uivj)i,j=1nAB = \pm (u u^\top)(v v^\top) = \pm u (u^\top v) v^\top = \pm u \cdot \langle u, v \rangle \cdot v^\top = \pm \langle u, v \rangle \cdot \bigl( u_i v_j \bigr)_{i,j=1}^{n} and similarly BA=±u,v(viuj)i,j=1n.BA = \pm \langle u, v \rangle \cdot \bigl( v_i u_j \bigr)_{i,j=1}^{n}. Since n=2025n = 2025 is odd, it follows that u,v0\langle u, v \rangle \ne 0. The first columns of the matrices uv=(uivj)i,j=1nu v^\top = \bigl( u_i v_j \bigr)_{i,j=1}^{n} and vu=(viuj)i,j=1nv u^\top = \bigl( v_i u_j \bigr)_{i,j=1}^{n} are uu and vv, respectively. Hence, AB=BAAB = BA if and only if u=vu = v; in other words, if A=±BA = \pm B.

For each ASA \in S, there are precisely two suitable matrices BSB \in S, so the probability that A,BA, B commute is 2S=12n1\dfrac{2}{|S|} = \dfrac{1}{2^{n-1}}.

How the field did

contestants scored
425
average (of 10)
6.32
solved (≥ 80%)
56.7%
near-0 (≤ 10%)
21.6%
discrimination
0.54

Score distribution (field cohort)

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

Similar problems

IMC 2007 · Day 1 · P2easyavg 7.0/10 · solved 55% · near-0 7% · disc 0.49
IMC 2017 · Day 1 · P1easyavg 6.9/10 · solved 59% · near-0 10% · disc 0.35
IMC 2021 · Day 2 · P5easyavg 5.8/10 · solved 55% · near-0 31% · disc 0.55
IMC 2017 · Day 2 · P8easyavg 6.2/10 · solved 52% · near-0 28% · disc 0.50