Positivity has a spectral consequence
Consider a real matrix
We will prove directly that it has a positive eigenvector for a real eigenvalue strictly larger in magnitude than its other eigenvalue. The proof is specific to dimension two; it does not assume a higher-dimensional Perron theorem.
Definition 7.1.1
(Dominant eigenvalue and spectral gap)
A real eigenvalue of a matrix is dominant when the other eigenvalue satisfies . The ratio is the factor by which the non-dominant component shrinks, relative to the dominant one, at each multiplication.
The two eigenvalues
Theorem 7.1.2
(Dominant eigenvalue of a positive matrix)
Let . Then has two distinct real eigenvalues with , and a right eigenvector for with strictly positive entries.
Proof
(Quadratic formula, then a positive eigenvector)
The characteristic equation has roots
Because , the square root is strictly greater than . Thus the roots are distinct and real, with
Also , so . Together these inequalities prove , including when is negative or zero.
A right eigenvector for is
whose entries are strictly positive. The first row verifies the eigenvalue equation directly: . The second row, , is the characteristic equation . A positive left eigenvector, viewed as a column, is , satisfying .
Proof X-ray
Why the negative eigenvalue case needs a separate check
The proof gets distinctness from : without it the square root could vanish, as for a diagonal matrix with equal entries. Positivity of the eigenvector costs nothing for the first entry, ; the second entry is positive because , again a consequence of .
The inequality compares signed numbers; alone it would not rule out a negative eigenvalue of larger magnitude. The positive trace gives , which supplies the missing half of the magnitude comparison. Here, and in the positivity , the proof uses positivity of the diagonal entries and not only of and .
Normalize repeated multiplication
Start with a vector whose entries are strictly positive and sum to 1, and define
Theorem 7.1.3
(Convergence of the normalized iteration)
With as above and strictly positive with entries summing to 1, the iterates satisfy and
Proof
(Expand in the eigenbasis and divide by )
Every denominator is positive because and have positive entries. Induction gives . Since the eigenvalues are distinct, their eigenvectors form a basis; write . The coefficient is positive. Indeed, and , while : applying to in two ways gives . Hence .
Dividing numerator and denominator by , the term involving contains . The denominator tends to the positive number , so .
The ratio controls the decaying component; a negative can make that component alternate in sign. This conclusion uses strict positivity of the matrix and the stated positive start.
Worked examples
Worked example 7.1.4
(A matrix with eigenvalues 5 and 2)
Take . Its eigenvalues are 5 and 2, with eigenvectors and ; the formula gives . From ,
Continuing, , so . The normalized iterates converge to ; the non-dominant eigencomponent is multiplied relative to the dominant one by per unnormalized power. Here , and , which agrees with after dividing by .
Worked example 7.1.5
(A negative second eigenvalue)
Take . The characteristic polynomial is , so , and . Here , and the limit is . Start with . Then and . Next , so . The error changes sign at each step, because the component along carries the factor , yet its size shrinks.
Worked example 7.1.6
(The smaller eigenvalue is zero)
Take , all entries positive. The eigenvalues are , so and still holds, with . For any positive start with entries summing to , , so already equals the limit: the decaying component dies in one step.
Counterexample mode
Nonnegative is weaker than strictly positive
Claim: the conclusion holds for every matrix with nonnegative entries. The identity matrix has nonnegative entries but a repeated eigenvalue 1 and no strict spectral gap. Its normalized iteration leaves every start unchanged: from it stays at , and from it stays at , so there is no single limiting direction for all starts. The permutation matrix is a second example: its eigenvalues are and , so , and from the iterates alternate between and without converging. Repair: require every entry strictly positive; zeros in the matrix cannot simply be allowed without rechecking the conclusion.
Exercises and solutions
Quick check
(In the first worked example, verify .)
Solution · Answer
.
Quick check
(Why is positive in the proof?)
Solution · Answer
It exceeds , and both diagonal entries are positive.
Quick check
(What is the sum of the entries of each normalized iterate?)
Solution · Answer
It is 1, by the defining denominator : the entries of sum to .
Quick check
(If is negative, does the proof require it to be positive?)
Solution · Answer
No. The proof establishes , and hence the ratio's powers tend to zero whatever the sign of .
Quick check
(Find and a positive right eigenvector for when .)
Solution · Answer
. The eigenvector is ; indeed .