48

Why is that if every row of a matrix sums to $1$ then the rows of its inverse matrix sum to $1$ too?

For example, consider

$$A=\begin{pmatrix} 1/3 & 2/3 \\ 3/4 & 1/4 \end{pmatrix}$$

then its inverse is

$$A^{-1}=\begin{pmatrix} -3/5 & 8/5 \\ 9/5 & -4/5 \end{pmatrix},$$

which satisfies the condition. Is it true for every such matrix?

Rodrigo de Azevedo
  • 23,223
Garmekain
  • 3,204

1 Answers1

108

Let $v = (1, 1, \ldots , 1)'$ be a column vector of all $1$s. Then the rows of $A$ adding to $1$ is equivalent to saying $Av = v$. So when $A$ is invertible, we will have $$A^{-1}v = A^{-1}Av = v$$ Thus $A^{-1}$ has rows summing to $1$ as well. (Note that $A$ will not always be invertible.)

Suhas Vijaykumar
  • 1,646
  • 5
    To extend the assertion we can say that property of row sum equal 1 holds also for any matrix $A^k$ ($A$ invertible) when $k$ is an integer $(...,-2,-1,0,1,2, ..)$ – Widawensen Jan 25 '18 at 10:33
  • 2
    The name of such a matrix is a right stochastic matrix (https://en.wikipedia.org/wiki/Stochastic_matrix), and it has a few more interesting properties. – nojka_kruva Jan 25 '18 at 16:40
  • @VictorMay The condition here is more general than that of stochastic. These entries don't need to be non-negative. – John Coleman Jan 25 '18 at 17:25
  • A right stochastic matrix is not required to have nonnegative values. – nojka_kruva Jan 25 '18 at 21:51
  • 3
    @VictorMay This is probably one of those cases where different definitions can be found in different sources, e.g.: https://en.wiktionary.org/wiki/right_stochastic_matrix . Some authors require it, some don't. I suspect that most do (since it is frequently defined in the context of Markov chains where the elements are probabilities) but haven't done any systematic survey. I was mistaken in assuming that the definition that I was familiar with was universal. – John Coleman Jan 25 '18 at 23:27
  • @JohnColeman I just scoured the shelf I have beside my computer. I have about 2 dozen books that define a "stochastic matrix" and not a single one that allows nonnegative entries. Perhaps a general definition is given by some authors, but the reference would be in the minority, and, in my opinion, very poorly named. – David Jan 30 '18 at 20:34
  • 1
    @David That meshes with my impression. Perhaps I was trying too hard to be diplomatic. I did find a couple of definitions online which didn't explicitly contain the phrase "nonnegative" (which caused me to think that Victor's use is not unprecedented) but I suspect now that those cases were just somewhat sloppy and assumed nonnegative as being clear from context. Nonnegative is both natural and at least nearly universal in the definition. – John Coleman Jan 30 '18 at 20:45
  • @JohnColeman understandable, i just wanted to chime in to make it a bit clearer (as it is a common term in my field). I too am a fan of diplomacy, but I'm also a fan of minimizing the future confusion of other readers (particularly as this post has gotten quite popular). – David Jan 30 '18 at 20:49