A symmetric matrix can have positive diagonal entries without being positive definite. Positivity is a statement about every nonzero direction, including combinations of coordinates. Cholesky factorization turns that global condition into positive pivots. The leading-minor criterion then gives a test in terms of determinants. To understand the test, we must show both why these pivots exist and how each determinant records their product.
A block identity
Definition 5.2.1
(Positive definite, Cholesky factor and leading minors)
A real symmetric matrix is positive definite if for every nonzero . A Cholesky factor of is a lower triangular matrix with positive diagonal entries such that . The th leading minor is the determinant of the upper-left block of , with .
Let be a real symmetric matrix, , and write
where . The matrix is the Schur complement of . For , completing the square yields
Lemma 5.2.2
(Reduction by the first pivot)
With , and as above, is positive definite if and only if and is positive definite.
Proof
(Reduction by completing the square)
If is positive definite, first take to get . Then for any nonzero , choose . This is nonzero and the square vanishes, so . Thus is symmetric positive definite. Conversely, if and is positive definite, the displayed sum is positive for every nonzero : when the second term is positive, and when , the first is positive because .
Proof X-ray
Why the square must vanish
The non-obvious move is the choice : it makes the first term of the identity zero, so reduces to and a statement about becomes a statement about the smaller matrix . The hypothesis allows the division, and gives the first term a positive coefficient in the converse. The positivity cannot be dropped: for we have and , which is positive definite, yet . The case in the converse is where is needed, because then the second term gives no positivity.
That converse matters: positivity is not just inherited by a smaller matrix; the smaller problem and the first pivot together completely characterize the original one. No division by is legitimate before proving it nonzero.
Existence and uniqueness of Cholesky
Theorem 5.2.3
(Cholesky factorization)
Every real symmetric positive-definite matrix has a unique lower triangular factor with positive diagonal satisfying . Conversely, if with lower triangular with positive diagonal, then is positive definite.
Proof
(Induction on the size via the Schur complement)
We prove existence and uniqueness by induction. For , the factor is and the positive square root is unique. For the induction step, the block argument gives and positive-definite . By induction for a unique lower triangular with positive diagonal. Set
Block multiplication gives the upper-left entry , the lower-left block , and the lower-right block . Thus . For uniqueness, any positive-diagonal lower triangular factor must have first entry ; its lower first column is then forced to be . Its remaining block must factor , so induction forces that block to equal . Positivity of the diagonal removes the sign ambiguity in the square roots.
Conversely, if with lower triangular and positive diagonal, then is invertible. For , , so . This proves the equivalence, not merely a procedure that happens to work in examples.
Why leading minors work
Theorem 5.2.4
(Leading-minor criterion)
A real symmetric matrix is positive definite if and only if for . In that case the Cholesky factor satisfies .
Proof
(Minors as squared products of pivots)
If , its leading block is , where is the leading block of . Hence
For the converse, suppose all . The first pivot is positive. Apply the block elimination matrices with determinant one to each leading block: the resulting block diagonal matrix has blocks and the leading block of . Therefore that leading determinant of is . By induction is positive definite; the completed-square identity then makes positive definite. For the statement is immediate. Thus a real symmetric matrix is positive definite exactly when all its leading principal minors are positive.
As a practical consequence, . A zero or negative leading minor is a stop signal for positive-definite Cholesky, not permission to take an arbitrary square root. The condition of symmetry is essential to the stated test.
The word “leading” fixes the order: the first rows and first columns must be chosen together. An arbitrary principal minor uses the same set of row and column indices, but need not be leading. For a positive-definite matrix every principal submatrix is positive definite, because a nonzero vector on the selected coordinates can be extended by zeros. The criterion above is more economical: it needs only the nested leading blocks. Reordering coordinates changes those blocks, although it cannot change positive definiteness. This explains why different orderings can give different pivots while reaching the same conclusion.
What nonnegative leading minors do not prove
Counterexample mode
Weak inequalities on leading minors do not give semidefiniteness
Changing every strict inequality in the criterion to a weak inequality does not give a positive-semidefinite test. Take . Its leading minors are and , both nonnegative. Yet , so is not even positive semidefinite. A zero first pivot conceals what happens in later directions. Positive semidefinite matrices can have singular triangular factorizations, but the positive-diagonal construction and its uniqueness proof above do not apply without adjustment.
Common mistakes
Common mistake
(Positive diagonal entries are not enough)
The symmetric matrix has positive diagonal, but for we get . Consistently, . Positivity along the coordinate axes says nothing about combined directions.
Common mistake
(A positive determinant alone does not prove positive definiteness)
For the determinant is , yet . The criterion needs every leading minor to be positive; here already fails.
Worked examples
Worked example 5.2.5
(Construct and verify a factor)
Let . Find its Cholesky factor and certify positive definiteness.
The first pivot is , so the first column of is . The Schur complement is . Its first Cholesky column is ; the last diagonal is . Hence
Direct multiplication gives . All diagonal entries are positive, so is invertible and for . The leading minors give an independent determinant check.
Worked example 5.2.6
(A parameter and an explicit failure direction)
For , determine precisely when it is positive definite. When it is not, give a nonzero vector with nonpositive quadratic form.
Its first leading minor is and its determinant is , so the criterion gives . Equivalently, the Schur complement is . If , take ; direct calculation gives . At the matrix is positive semidefinite but singular; below that value the same vector makes the form negative. This distinguishes failure of strict positivity from the stronger conclusion of indefiniteness.
Worked example 5.2.7
(Failure at the second pivot)
Test .
The first pivot is and , so . Its first diagonal entry is negative, so the Cholesky procedure stops at the second step. The leading minors are , , , consistent with this. An explicit witness is , for which . The matrix is not positive definite even though .
Quick checks
Quick check
(Does the leading-minor test certify ? Give a nonzero direction that shows what happens.)
Compute .
Solution · Answer
No. but , so is not positive definite. For , , so the quadratic form vanishes on a nonzero vector.
Quick check
(A positive-definite matrix has leading minors . What are the diagonal entries of its Cholesky factor?)
Use with .
Solution · Answer
, , . Taking positive square roots, the diagonal is .
Quick check
(For with , when is positive definite?)
The Schur complement of is a number.
Solution · Answer
The Schur complement is . By the reduction lemma, is positive definite exactly when , that is, (equivalently ).