A,B,C are n $\times$ n symmetric and idempotent matrices. The question is:

If $$ \textbf{A = B + C } $$ then show

$$ \textbf{BC = 0 } $$

I'm not sure where to start?

  • 5
    $\begingroup$ Have you tried squaring both sides of the equality ? $\endgroup$ – Joel Cohen Feb 6 at 4:44

Here's my expansion of Brian's argument:

$A^{2}=A$, $B^{2}=B$, $C^{2}=C$, hence

$A^{2}=(B+C)^{2}\rightarrow BC+CB=0 \implies BC+(BC)^{T}=0$ Hence $BC$ is an anti-symmetric matrix, thus it has only imaginary eigenvalues.

$ BC+CB=0 \implies B(BC+CB)C=0 \implies BC + (BC)^{2}=0$,

If $BC\neq0$ then $BC + (BC)^{2}=0$ is the minimal-polynomial for $BC$, hence $-1$ is an eigenvalue of $BC$, which is a contradiction since $BC$ can have only imaginary eigenvalues


Recall that a matrix $M$ is idempotent if $M^2=M$.

Thus, since $A=B+C$ where $A$, $B$, and $C$ are idempotent implies that $$ B+C=(B+C)^2=B^2+C^2+BC+CB=B+C+BC+CB $$ Simplifying this equation gives $$ 0=BC+CB=BC+(B^\top C^\top)^\top $$ How might we proceed from here?

  • $\begingroup$ At this point, since we have, BC = -(B'C')', can we not argue that BC has to equal zero for this to be true? $\endgroup$ – user1992460 Feb 6 at 5:15
  • $\begingroup$ Note that for symmetric matrices $S$, we have $S = S^T$. $\endgroup$ – Trevor Kafka Feb 6 at 5:17
  • $\begingroup$ Just from the symmetric argument, neither $BC=CB$ must be true, nor $BC$ is necessarily symmetric as answered here math.stackexchange.com/questions/874469/… $\endgroup$ – cr001 Feb 6 at 5:21
  • 2
    $\begingroup$ I have tried to extend your argument to get to a contradiction; take a look. Thanks $\endgroup$ – piyush_sao Feb 6 at 5:44

We can also proceed from the step $0=BC+CB$ this way:

Multiply both sides by $B$ we have


but $I+B$ is full rank matrix (it doesn't have zero eigenvalue because for $B+I$ all eigenvalues: - $0,1$ for idempotent matrix $B$, were shifted by $+1$ with reference to eig. of $B$) so $BC$ annihilates full rank matrix,

hence $BC$ has to be zero matrix.


Note that the claim is not true over a field of characteristic $2$: consider the matrices $B=C=\mathbf 1$ and $A=\mathbf0$.

This counterexample shows that you're not going to be able to prove the claim simply by performing the sort of algebraic manipulations that work over all fields, such as $(XY)^T=Y^TX^T$ and so on. You're going to need some argument that involves the base field. For example, see zimbra314's answer in terms of complex eigenvalues.

  • 1
    $\begingroup$ How about using the spectrum decomposition in some way? $\endgroup$ – user1992460 Feb 6 at 6:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.