From a defining equation to a calculation
The simple-extension theorem says that an algebraic element has a power basis determined by its minimal polynomial. In an actual calculation, however, the element whose minimal polynomial we want may be a complicated expression in a different generator. We might know a basis for the surrounding field without yet knowing a minimal polynomial for that expression.
Linear algebra reverses the usual direction. Write successive powers of the chosen element in the basis we already know, and stop at the first dependence. The resulting relation is minimal because every earlier power was independent. This supplies both an equation and a proof of its minimality; finding an equation alone supplies only the first of these.
We will carry out this procedure in the real field , where is the positive fourth root of . For , we will obtain its minimal polynomial, recover from , and compute . A second element of the same field will show why an element's degree can be smaller than the degree of the surrounding field.
What must be known before computing
Let be a finite field extension of degree , and suppose is a known -basis. For , write
The coordinates are unique because is a basis. A list that merely spans is not enough for the coordinate arguments below unless it is first reduced to a basis. The basis need not begin with ; the coordinate column of must be computed in the chosen basis too.
To multiply coordinate columns, we need to know the products in the same basis. In a simple extension with a known minimal polynomial, its defining relation reduces high powers of and provides these products. More generally, known coefficients in
allow multiplication by expansion. Addition and multiplication alone therefore let us form .
All linear dependence here is over . Approximating real roots and using a numerical tolerance is not a proof of an exact relation. This is an algebraic method assuming that the required arithmetic and linear equations in can be performed; it is not a claim that an arbitrary abstract field comes with an effective computer representation.
The first dependence gives the minimal polynomial
Theorem 1.3.1
(Recovering the minimal polynomial from powers)
Let , where . There is a least integer for which are linearly dependent over , and . There are unique coefficients with
The minimal polynomial of over is
Furthermore, is an -basis of , and generates exactly when .
Proof
(Why stopping at the first relation proves minimality)
The vector is nonzero, while any vectors in an -dimensional space are dependent. Thus the least exists and satisfies . By minimality, are independent. In a nontrivial relation involving also , its coefficient must be nonzero; otherwise the earlier powers would already be dependent. Divide by that coefficient and isolate . Independence of the preceding powers proves uniqueness of the .
The displayed monic polynomial vanishes at , and no nonzero polynomial of degree less than does, since such a polynomial would give a relation among the preceding powers. It is therefore the monic polynomial of least degree vanishing at . It is irreducible: if it were a product of two positive-degree polynomials over , evaluation at would give a product equal to zero in the field . One factor would vanish at , contradicting the least degree.
The simple-extension theorem now gives with the stated basis. If , this subfield is an -subspace of of the same finite dimension, hence equals . Conversely, if , their dimensions are equal and .
In coordinates, append one column at a time to
The first new column in the span of its predecessors is . Solve the corresponding linear system for . Reducing the powers in and row reducing the columns over are separate steps: the first computes the data, and the second detects the relation.
Changing the known basis cannot change the answer. Under a change of basis, every coordinate column is multiplied by the same invertible matrix over . A linear combination of the new columns is zero exactly when the corresponding combination of the old columns is zero: multiply by the inverse matrix for the reverse implication. Thus both the first dependence and its coefficients are unchanged. A convenient basis can simplify the arithmetic without changing the minimal polynomial.
The boundary cases also fit. If , the first relation is , so and . In particular, gives . There is no degree-zero relation, because the single vector is nonzero.
Proof X-ray
An equation needs a lower-degree argument
An equation proves only that divides . The independent columns for rule out every smaller-degree equation at once. Both pieces are needed to identify the minimal polynomial. The degree of the surrounding field is only an upper bound until independence has been established.
A complete quartic computation
Worked example 1.3.2
(A new generator inside a fourth-root field)
Let and . Eisenstein's criterion at proves that is irreducible over . Thus is a basis and .
Set . Multiplication followed by reduction using gives
For example, before reduction the cube is ; replacing the last three powers gives the middle line. The coordinate columns for the first four powers form
Expanding along the first column gives
Hence are independent over . Comparing the displayed coordinates of with those of and gives
This is the first dependence. Therefore
Its irreducibility follows from the independent columns and the theorem. Merely checking that it has no rational root would not rule out a factorization into two quadratics.
This computation used no list of conjugates and no assumption that is normal. The coordinates are taken in the actual field in which the element lies. The same method applies in positive characteristic and to inseparable elements, provided a valid basis is known.
Membership is another linear system
Once the first dependence occurs at , an element belongs to exactly when
has a solution with . If it has a solution, independence makes it unique, and it explicitly writes as a polynomial in . If it has no solution, is outside the subfield. It is enough to use polynomials here because is algebraic: by the simple-extension theorem.
Worked example 1.3.3
(Recovering the original fourth root)
For the quartic example, solve . The equations are
The last equation gives ; substituting into the middle two gives and . Finally and . Consequently
The formula is an explicit witness to , in addition to the earlier dimension proof. Its denominator is legitimate because the base field here is ; the computation must not be transferred unchanged to characteristic three.
Worked example 1.3.4
(A proper subfield in the same ambient field)
Keep but take . The vectors are independent in , while . Thus and has basis . Its coordinate vectors are exactly with .
The element has coordinates and therefore does not belong to this subfield. The tower law gives . Running the procedure in a degree-four field did not force to have degree four.
Computing inverses without guessing
Proposition 1.3.5
(Read an inverse from the minimal polynomial)
Suppose has minimal polynomial
Then , and
For , the numerator is ; there are no further terms.
Proof
(Separate the constant term)
If , write . Evaluation gives . Since and is a field, , contrary to minimality. Hence . Now can be written
Dividing by the nonzero scalar proves the formula. The excluded element has minimal polynomial and no inverse.
For the quartic element above, the formula gives
Multiplying by and replacing by makes the product equal to , so the answer can be checked directly.
There is also an inverse calculation using the original basis. Form the matrix whose th column is . Then . For , multiplication by is injective on the field and therefore bijective on its finite-dimensional -vector space. Thus has a unique solution, the coordinates of . This method is useful when an inverse is needed before the minimal polynomial has been computed.
Concept lensAlgebraic
The power columns as a Krylov sequence
The columns used to find are , so the procedure computes the first linear relation among the vectors . Its answer is also the minimal polynomial of the matrix : for any , , so forces , which in turn makes . Thus is the minimal polynomial of , while, as the common mistake below notes, the characteristic polynomial of can have larger degree.
Worked example 1.3.6
(Inverting one plus the fourth root)
For in the same basis,
Solving gives , , , . The solution is , so
The product is . This is the same inverse that a polynomial Bézout identity would produce; coordinates offer a different way to find it.
The characteristic polynomial of a multiplication matrix
The matrix carries more information than the minimal polynomial. Its characteristic polynomial always has degree , whatever the degree of , and it is a power of .
Theorem 1.3.7
(The characteristic polynomial is a power of the minimal polynomial)
Let , let have minimal polynomial of degree over , and put , so that . Then
In particular, does not depend on the chosen basis, and exactly when generates over .
Proof
(Choose a basis adapted to the tower)
A change of basis replaces by , and . So we may compute in any basis.
Let be a basis of over . By the proof of the tower law, the products with and form an -basis of . Order it as . Multiplication by maps the span of into itself: it sends to for , and sends to , using the first-dependence relation from the main theorem. Hence, in this basis, is block diagonal with equal blocks . The block has ones directly below the diagonal, the coefficients in its last column, and zeros elsewhere; it is the companion matrix of . The determinant of a block-diagonal matrix is the product of the determinants of its blocks, so .
It remains to show . The block for is the matrix of multiplication by on in the power basis . Write , a monic polynomial of degree . By the Cayley–Hamilton theorem, . Since is the matrix of multiplication by on , applying it to the coordinate column of gives . Therefore divides ; both are monic of degree , so .
Finally, exactly when , which is the last assertion.
Worked example 1.3.8
(A characteristic polynomial that is a square)
In with and basis , let . The products , , and give
Reordering the basis as , which is the adapted basis of the proof with and , turns into two copies of the companion matrix of . Hence
as the theorem predicts with . The characteristic polynomial has degree four, while the minimal polynomial has degree two. The trace of is and its determinant is ; these are the trace and norm of studied in the trace-and-norm unit.
The theorem gives a second method for finding : compute and extract the monic irreducible polynomial whose th power it is. The first-dependence method is usually shorter, because it stops as soon as the relation appears and needs no determinant of size .
Why the field hypothesis cannot be omitted
The irreducibility step in the theorem used one property of that a general finite-dimensional algebra lacks: a product of two elements vanishes only when one factor vanishes.
Counterexample mode
A first dependence in an algebra that is not a field
Claim to test: in every finite-dimensional commutative -algebra, the first dependent power of an element gives an irreducible polynomial. Consider the algebra , and let be the class of . Division by the monic quadratic shows that is a -basis of this algebra. The first dependence is , so the least monic annihilating polynomial is . But this polynomial is reducible.
There is no contradiction with the theorem: is not a field. Its nonzero elements and have product zero. The irreducibility proof used the fact that a product of two elements of a field can vanish only when one factor vanishes. Linear dependence still finds a least annihilating polynomial in a finite-dimensional algebra, but it does not make that polynomial irreducible or make every nonzero element invertible. The repaired statement is the theorem as stated: the ambient object must be a field.
Common mistake
(Three objects that need separate justification)
A defining relation, a basis of the ambient field, and a minimal polynomial of the particular element are different data. A polynomial quotient is a field only after irreducibility is established. An ambient basis gives an upper bound on an element's degree, while the first dependent power determines its actual degree. The characteristic polynomial of a multiplication matrix is yet another object: by the theorem above it is , which has higher degree than unless generates .
Practice
In the first four questions, let be the positive fourth root of and use the basis of .
Quick check
(For , find its minimal polynomial and its inverse. Explain why the equation you find is minimal.)
Solution · Answer
We have . The vectors and are independent, because the coordinate of is . Thus the first relation has degree two and . Since , its inverse is .
Quick check
(Determine whether and belong to .)
Solution · Answer
Elements of this subfield have zero coefficients at and . Therefore belongs to it, while does not. Uniqueness of coordinates rules out cancellation by a different expression in .
Quick check
(Put . Find the coordinate columns of , deduce its minimal polynomial, and recover from .)
Solution · Answer
The columns are , , , and , since and . They are independent and , so . Also , explicitly showing that generates .
Quick check
(Check the formula for using only its quartic relation.)
Solution · Answer
The product is . Substitute to obtain .
Quick check
(Let with an indeterminate, and adjoin satisfying . Prove that the first-dependence method gives even though its derivative is zero.)
Solution · Answer
The element is not a square in . Otherwise nonzero polynomials would satisfy , giving , an impossibility. Hence has no root in and is irreducible. Thus is a basis of a degree-two field extension. Its two vectors are independent and is the first relation. The theorem gives . Separability was never assumed; the zero derivative means that this irreducible polynomial has a repeated root over an algebraic closure.
Quick check
(In , why does the nonzero element have no inverse? What specifically fails if one tries the inverse-matrix argument?)
Solution · Answer
The class is nonzero, but . An inverse for would force . Multiplication by is therefore not injective, so its matrix is singular. The step from nonzero multiplier to injective multiplication is valid in a field and fails in this algebra.
Exercise 1.3.9
(In with basis , use coordinates to find the minimal polynomial of over and to express as a polynomial in .)
Core
Hint(1)
Compute , and in the basis, test the first four columns with a determinant, then solve one linear system.
Solution(4)
- 1
Coordinates of the powers
, and . The columns of are , , and .
- 2
Independence of the first four powers
Key step
Expanding along the first column, the determinant equals . So are independent and has degree four.
- 3
The first dependence
has coordinates , the same as . Hence and ; in particular .
- 4
Recover the square root
Solving gives , , and , so , and . Thus , matching the computation in the introductory unit.
Quick check
(For in the quartic example, what is ? For , what are and ?)
Solution · Answer
Since generates , and . For , , and are independent, so . Here , so .
Quick check
(An element of a degree-six extension has . Find and .)
Solution · Answer
By the theorem, with monic irreducible. The polynomial is irreducible over , so unique factorization in forces and . Hence .