An operator built from a vector and a functional
Let be a vector space over a field , let , and let be linear. This construction connects vectors, dual functionals and operators without choosing coordinates.
Definition 7.1.1
(Rank-one operator)
For and a linear functional , define . Its image is contained in , so its rank is at most one; it is zero if either or is zero. Write .
Proposition 7.1.2
(Square of a rank-one operator)
For and , we have on every vector.
Proof
(Pull the scalar through the functional)
Linearity gives
The scalar , not the rank alone, controls the next calculation.
Inverting an update
Theorem 7.1.3
(Inverse of a rank-one update)
With and as above, is invertible if and only if in , and then
No finite-dimensionality is assumed.
Proof
(Multiplying out, and a kernel vector when )
Suppose . Multiply in either order and use : the coefficient of is , so both products equal . The scalar must be nonzero before division; in positive characteristic, a familiar integer can represent zero.
Suppose . Then and , since and in a field. Moreover , so has a nonzero kernel vector and cannot be invertible. Therefore the condition is necessary as well as sufficient.
Proof X-ray
Two halves with different engines
The sufficiency half is pure algebra. The proposed inverse is a polynomial in , so it commutes with , and the single expansion using shows that both compositions equal ; no theorem such as "a one-sided inverse is two-sided in finite dimensions" is silently invoked, which is why the result holds in every dimension. The only division is by , and this is the single place where the field enters. The necessity half uses a different idea, not an estimate but a witness: when the vector itself is sent to by . The hypothesis is not assumed but derived from . Dropping the condition therefore does not leave the formula slightly wrong; it removes invertibility altogether.
Worked examples
Worked example 7.1.4
(Inverting a rank-one update over the reals)
Take , and . Then and , so
For instance, gives ; substitution returns . Here is nonzero because the field is .
Worked example 7.1.5
(A nonzero nilpotent operator)
Now let but over . Then while . Hence but . The inverse formula becomes .
Worked example 7.1.6
(The singular case in characteristic three)
Over take , and . Then , so . Indeed , so is a nonzero kernel vector and is not invertible, although has rank one and .
Counterexample mode
The field affects invertibility
A tempting claim is that always makes a harmless nonzero denominator. In , take , and . Then and , so . Consequently and is not invertible. Dividing by the real number 2 would ignore the field.
Common mistake
(A zero value of does not mean )
In the nilpotent example above, yet . The value only controls ; the operator is zero only when or .
Exercises and solutions
Quick check
(For the real first example, solve .)
Use .
Solution · Answer
gives , and gives , so .
Quick check
(If , what does the inverse formula give?)
What is when ?
Solution · Answer
Here forces or , so and the formula gives .
Quick check
(In the nilpotent example, compute .)
Use .
Solution · Answer
, because .
Quick check
(For a nonzero finite-dimensional , take and a functional with . What is a nonzero kernel vector of ?)
Apply to .
Solution · Answer
The vector itself: .