uniolympiads Ελληνικά

Nilpotent and commuting matrices

When a matrix is nilpotent, what you gain when AB=BAAB = BA, and the trick det⁡(A2+B2)=∣det⁡(A+iB)∣2\det(A^2 + B^2) = |\det(A + iB)|^2.

Nilpotent matrices

N∈Mn(C)N \in \M_n(\C) is called nilpotent if Nk=0N^k = 0 for some kk. The following are equivalent:

  1. NN is nilpotent;
  2. all its eigenvalues are 00;
  3. χN(x)=xn\chi_N(x) = x^n, hence Nn=0N^n = 0 (the exponent nn always suffices);
  4. tr⁡(Nk)=0\tr(N^k) = 0 for k=1,…,nk = 1, \dots, n.

(4) ⇒ (2) comes from Newton's identities. If the non-zero eigenvalues were μ1,…,μr\mu_1, \dots, \mu_r with multiplicities mjm_j, then ∑jmjμjk=0\sum_j m_j \mu_j^k = 0 for k=1,…,rk = 1, \dots, r, and the Vandermonde determinant gives mj=0m_j = 0. It needs characteristic 00 (over Fp\F_p, IpI_p has tr⁡(Ik)=0\tr(I^k) = 0).

Inversion. If NN is nilpotent, I+NI + N is invertible with (I+N)−1=I−N+N2−⋯+(−1)n−1Nn−1(I + N)^{-1} = I - N + N^2 - \dots + (-1)^{n-1} N^{n-1}.

Commuting matrices

If AB=BAAB = BA in Mn(C)\M_n(\C):

  • Common eigenvector. Each eigenspace of AA is invariant under BB, so BB has an eigenvector inside it.
  • Simultaneous triangularisation: there is PP with P−1APP^{-1}AP and P−1BPP^{-1}BP upper triangular. Hence the eigenvalues of A+BA + B are λi+μi\lambda_i + \mu_i and those of ABAB are λiμi\lambda_i \mu_i, for a suitable matching.
  • If both are diagonalisable, they are simultaneously diagonalisable.
  • The binomial formulas hold: (A+B)m=∑k(mk)AkBm−k(A + B)^m = \sum_k \binom{m}{k} A^k B^{m-k}, and eA+B=eAeBe^{A + B} = e^A e^B.
  • The sum and the product of commuting nilpotent matrices are nilpotent.

The commutator [A,B]=AB−BA[A, B] = AB - BA always has trace 00. So AB−BA=IAB - BA = I is impossible (the trace would be nn). If moreover [A,B][A, B] commutes with AA, then [A,B][A, B] is nilpotent. Proof: with C=[A,B]C = [A, B] and CA=ACCA = AC, we get Ck=[A,BCk−1]C^k = [A, BC^{k-1}], so tr⁡(Ck)=0\tr(C^k) = 0 for every kk, and (4) applies.

Passing to the complex numbers

For real A,BA, B with AB=BAAB = BA:

A2+B2=(A+iB)(A−iB),det⁡(A2+B2)=det⁡(A+iB) det⁡(A+iB)‾=∣det⁡(A+iB)∣2≥0.A^2 + B^2 = (A + iB)(A - iB), \qquad \det(A^2 + B^2) = \det(A + iB)\,\overline{\det(A + iB)} = |\det(A + iB)|^2 \ge 0 .

Special case B=IB = I: det⁡(A2+I)≥0\det(A^2 + I) \ge 0. Without AB=BAAB = BA the factorisation fails, because (A+iB)(A−iB)=A2+B2+i(BA−AB)(A + iB)(A - iB) = A^2 + B^2 + i(BA - AB).

Other quick arguments for real matrices:

  • For odd nn, every A∈Mn(R)A \in \M_n(\R) has a real eigenvalue, because χA\chi_A has odd degree.
  • If A2=−IA^2 = -I in Mn(R)\M_n(\R), then (det⁡A)2=(−1)n(\det A)^2 = (-1)^n, so nn is even.

Pitfalls

  • Without AB=BAAB = BA, none of the above holds. (0100)\begin{pmatrix} 0 & 1 \\ 0 & 0 \end{pmatrix} and (0010)\begin{pmatrix} 0 & 0 \\ 1 & 0 \end{pmatrix} are nilpotent, but their sum has eigenvalues ±1\pm 1.
  • eA+B=eAeBe^{A+B} = e^A e^B needs commutativity. Don't write it without checking.
  • In (4) you need all k=1,…,nk = 1, \dots, n, not only tr⁡N=0\tr N = 0.

Further reading

  • Axler, Linear Algebra Done Right, 4th ed., §8A (Generalized Eigenvectors and Nilpotent Operators)
  • Axler, Linear Algebra Done Right, 4th ed., §5E (Commuting Operators)
  • Axler, Linear Algebra Done Right, 4th ed., §1B (Exercise 8: the complexification V_C)
  • Axler, Linear Algebra Done Right, 4th ed., §3B (Exercise 33: the complexification T_C)
  • Axler, Linear Algebra Done Right, 4th ed., §5A (Exercise 17: real eigenvalues of T and T_C)

Contest problems using these tools

No problems with these tools yet.