uniolympiads Ελληνικά

IMC 2015, Day 2, Problem 9

An n×nn \times n complex matrix AA is called t-normal if AAt=AtAAA^t = A^tA where AtA^t is the transpose of AA. For each nn, determine the maximum dimension of a linear space of complex n×nn \times n matrices consisting of t-normal matrices.

Original from the official PDFAn n × n complex matrix A is called t-normal if AA t = A t A where A t is the transpose of A . For each n , determine the maximum dimension of a linear space of complex n × n matrices consisting of t-normal matrices. Problem 9.

This transcription was compared symbol by symbol with the original, but not yet by a person.

Time on this problem 0:00

Work for at least 45 minutes before opening the solution.

Official solution

(Proposed by Shachar Carmeli, Weizmann Institute of Science)

Solution.

Answer: The maximum dimension of such a space is n(n+1)2\frac{n(n+1)}{2}.

The number n(n+1)2\frac{n(n+1)}{2} can be achieved, for example the symmetric matrices are obviously t-normal and they form a linear space with dimension n(n+1)2\frac{n(n+1)}{2}. We shall show that this is the maximal possible dimension.

Let MnM_n denote the space of n×nn\times n complex matrices, let Sn⊂MnS_n\subset M_n be the subspace of all symmetric matrices and let An⊂MnA_n\subset M_n be the subspace of all anti-symmetric matrices, i.e. matrices AA for which At=−AA^t = -A.

Let V⊂MnV\subset M_n be a linear subspace consisting of t-normal matrices. We have to show that dim⁡(V)≤dim⁡(Sn)\dim(V)\le\dim(S_n). Let π:V→Sn\pi: V\to S_n denote the linear map π(A)=A+At\pi(A) = A+A^t. We have

dim⁡(V)=dim⁡(Ker (π))+dim⁡(Im (π))\dim(V) = \dim(\mathrm{Ker}\,(\pi)) + \dim(\mathrm{Im}\,(\pi))
so we have to prove that dim⁡(Ker (π))+dim⁡(Im (π))≤dim⁡(Sn)\dim(\mathrm{Ker}\,(\pi)) + \dim(\mathrm{Im}\,(\pi))\le\dim(S_n). Notice that Ker (π)⊆An\mathrm{Ker}\,(\pi)\subseteq A_n.

We claim that for every A∈Ker (π)A\in\mathrm{Ker}\,(\pi) and B∈VB\in V, Aπ(B)=π(B)AA\pi(B) = \pi(B)A. In other words, Ker (π)\mathrm{Ker}\,(\pi) and Im (π)\mathrm{Im}\,(\pi) commute. Indeed, if A,B∈VA,B\in V and A=−AtA = -A^t then

(A+B)(A+B)t=(A+B)t(A+B)⇔⇔AAt+ABt+BAt+BBt=AtA+AtB+BtA+BtB⇔⇔ABt−BA=−AB+BtA⇔A(B+Bt)=(B+Bt)A⇔⇔Aπ(B)=π(B)A.\begin{gathered} (A+B)(A+B)^t = (A+B)^t(A+B) \Leftrightarrow \\ \Leftrightarrow AA^t + AB^t + BA^t + BB^t = A^tA + A^tB + B^tA + B^tB \Leftrightarrow \\ \Leftrightarrow AB^t - BA = -AB + B^tA \Leftrightarrow A(B+B^t) = (B+B^t)A \Leftrightarrow \\ \Leftrightarrow A\pi(B) = \pi(B)A. \end{gathered}

Our bound on the dimension on VV follows from the following lemma:

Lemma. Let X⊆SnX\subseteq S_n and Y⊆AnY\subseteq A_n be linear subspaces such that every element of XX commutes with every element of YY. Then

dim⁡(X)+dim⁡(Y)≤dim⁡(Sn)\dim(X) + \dim(Y)\le\dim(S_n)

Proof. Without loss of generality we may assume X=ZSn(Y):={x∈Sn:xy=yx∀y∈Y}X = Z_{S_n}(Y) := \{x\in S_n : xy = yx \quad \forall y\in Y\}. Define the bilinear map B:Sn×An→CB: S_n\times A_n\to\mathbb{C} by B(x,y)=tr(d[x,y])B(x,y) = \mathrm{tr}(\mathrm{d}[\mathrm{x},\mathrm{y}]) where [x,y]=xy−yx[x,y] = xy-yx and d=diag(1,…,n)d = diag(1,\ldots,n) is the matrix with diagonal elements 1,…,n1,\ldots,n and zeros off the diagonal. Clearly B(X,Y)={0}B(X,Y) = \{0\}. Furthermore, if y∈Yy\in Y satisfies that B(x,y)=0B(x,y) = 0 for all x∈Snx\in S_n then tr(d[x,y])=−tr([d,x],y])=0\mathrm{tr}(\mathrm{d}[\mathrm{x},\mathrm{y}]) = -\mathrm{tr}([\mathrm{d},\mathrm{x}],\mathrm{y}]) = 0 for every x∈Snx\in S_n.

We claim that {[d,x]:x∈Sn}=An\{[d,x] : x\in S_n\} = A_n. Let EijE_i^j denote the matrix with 1 in the entry (i,j)(i,j) and 0 in all other entries. Then a direct computation shows that [d,Eij]=(j−i)Eij[d,E_i^j] = (j-i)E_i^j and therefore [d,Eij+Eji]=(j−i)(Eij−Eji)[d,E_i^j+E_j^i] = (j-i)(E_i^j-E_j^i) and the collection {(j−i)(Eij−Eji)}1≤i<j≤n\{(j-i)(E_i^j-E_j^i)\}_{1\le i<j\le n} span AnA_n for i≠ji\ne j.

It follows that if B(x,y)=0B(x,y) = 0 for all x∈Snx\in S_n then tr(yz)=0\mathrm{tr}(\mathrm{yz}) = 0 for every z∈Anz\in A_n. But then, taking z=yˉz = \bar{y}, where yˉ\bar{y} is the entry-wise complex conjugate of yy, we get 0=tr(yyˉ)=−tr(yyˉt)0 = \mathrm{tr}(\mathrm{y}\bar{\mathrm{y}}) = -\mathrm{tr}(\mathrm{y}\bar{\mathrm{y}}^{\mathrm{t}}) which is the sum of squares of all the entries of yy. This means that y=0y = 0.

It follows that if y1,…,yk∈Yy_1,\ldots,y_k\in Y are linearly independent then the equations

B(x,y1)=0,…,B(x,yk)=0B(x,y_1) = 0, \quad \ldots, \quad B(x,y_k) = 0
are linearly independent as linear equations in xx, otherwise there are a1,…,aka_1,\ldots,a_k such that B(x,a1y1+…+akyk)=0B(x,a_1y_1+\ldots+a_ky_k) = 0 for every x∈Snx\in S_n, a contradiction to the observation above. Since the solution of kk linearly independent linear equations is of codimension kk,
dim⁡({x∈Sn:[x,yi]=0, for i=1,..,k})≤≤dim⁡(x∈Sn:B(x,yi)=0 for i=1,…,k)=dim⁡(Sn)−k.\begin{gathered} \dim(\{x\in S_n : [x,y_i] = 0, \text{ for } i = 1,..,k\})\le \\ \le\dim(x\in S_n : B(x,y_i) = 0 \text{ for } i = 1,\ldots,k) = \dim(S_n) - k. \end{gathered}
The lemma follows by taking y1,…,yky_1,\ldots,y_k to be a basis of YY.

Since Ker (π)\mathrm{Ker}\,(\pi) and Im (π)\mathrm{Im}\,(\pi) commute, by the lemma we deduce that

dim⁡(V)=dim⁡(Ker (π))+dim⁡(Im (π))≤dim⁡(Sn)=n(n+1)2.\dim(V) = \dim(\mathrm{Ker}\,(\pi)) + \dim(\mathrm{Im}\,(\pi))\le\dim(S_n) = \frac{n(n+1)}{2}.

Transcribed from the official solutions, pages 3–4. This transcription has not been checked a second time; if in doubt, look at the original.

Archive problem: the statement is the official text (transcribed, with the original above). Subject classification and difficulty were estimated automatically; hints and a worked solution are added over time.

How did it go?