A finite argument inside an infinite field
Definition 1.6.1
(Algebraic and finite extensions)
A field extension is algebraic if every element of satisfies a nonzero polynomial in . It is finite if its dimension as a -vector space is finite.
Finite extensions are algebraic, but algebraic extensions need not be finite. This distinction matters when a proof passes through an algebraic closure or adjoins infinitely many roots: multiplying possibly infinite degrees does not supply the finite witness needed to prove an element algebraic.
The useful observation is local. A single polynomial has only finitely many coefficients, and each rational expression in adjoined elements uses only finitely many of them. These finite pieces let the tower law do the work.
Theorem 1.6.2
(Finite generators and algebraicity)
Let be a field extension and let be algebraic over . Then is finite. Conversely, every element of a finite extension of is algebraic over .
Proof
(One adjunction at a time)
Put and . The nonzero polynomial over that vanishes at also belongs to . Hence is algebraic over , and the simple-extension theorem gives . The tower law gives .
Conversely, if , the vectors are linearly dependent over for any . Their nontrivial linear relation is a nonzero polynomial over vanishing at .
Concept lensAlgebraic
Algebraicity as finite dimension
For , three conditions are equivalent: is algebraic over ; the ring of polynomial expressions in is finite-dimensional over ; and is a field. A nonzero annihilating polynomial of degree makes span . If is finite-dimensional, multiplication by a nonzero element is injective and hence surjective, so inverses exist. If is a field and , then for some polynomial , and is a nonzero polynomial vanishing at . The injectivity step reappears in the compositum proof below.
Theorem 1.6.3
(Transitivity of algebraicity)
If , is algebraic and is algebraic, then is algebraic. No extension in the statement need be finite.
Proof
(Capture the coefficients)
Take . Choose a nonzero polynomial with as a root. Set . Each coefficient is algebraic over , so the preceding theorem makes finite. The same nonzero polynomial shows is algebraic over , so is finite. By the tower law is finite. Therefore is algebraic over . As was arbitrary, is algebraic.
Proof X-ray
Where finiteness enters
The whole intermediate field need not be finite over . Only the field generated by the finitely many coefficients of one equation must have finite degree over . This finite-degree subextension may change when changes; no uniform degree bound is asserted.
Reading the coefficient argument carefully
It helps to separate three fields in the proof. The coefficients are in , not necessarily in . Their generated field lies inside , and the particular element may lie outside . The equation has exactly the same coefficients when read over , and it remains nonzero: changing the coefficient field does not turn a nonzero coefficient into zero. This is why we may adjoin to even though no finite-degree statement has been made about all of .
There is also a quantitative form for a particular equation. If is annihilated by a nonzero polynomial over of degree at most with , and the equation for has degree , then
At each step, the minimal polynomial over the larger preceding field divides the given annihilating polynomial, so its degree is at most the displayed bound. The resulting product may be very wasteful: coefficients may repeat, lie in fields already generated, or have smaller minimal polynomials. The point is that it is finite for this one , not that a common bound works for all of .
Worked example 1.6.4
(Following coefficients through a tower)
Take , , and choose with . The equation has coefficients in , so . Since , the tower law yields . We can also eliminate the coefficient: implies
The displayed polynomial is a concrete rational equation, which is enough to establish algebraicity; it is not necessary here to prove it irreducible. The calculation makes the abstract proof visible: first confine the coefficients to a finite extension, then eliminate them.
Arbitrarily many generators
Let be a set of elements algebraic over . Then is algebraic, even if is infinite. Indeed, the union of the fields over finite is a field: two elements lie together in , where their sum, product and any needed inverse lie. It contains and , and is contained in every field containing them; hence it is . Each element lies in one finite extension and is therefore algebraic.
For the union claim, the case gives . If and , the union is finite, so both elements lie in the same field . This common field contains their sum and product; when it also contains . Thus the union is closed under the field operations. This argument is stronger than merely observing that each generator is algebraic: an element of can be a complicated quotient, but it still uses only finitely many generators.
Composita and degree bounds
Suppose and contain and sit in the same overfield . Their compositum is the smallest subfield of containing both. It is : one adjoins all elements of to . The common overfield matters because the notation records how these particular copies of the fields are situated.
Theorem 1.6.5
(A finite compositum bound)
If , then . In particular, if is also finite, then is finite and .
Proof
(Spanning is enough, but independence can be lost)
Let be a -basis of , and let be their -linear span in . It contains and , since . Products of basis elements lie in , and their -linear expansions show that is closed under multiplication. If is nonzero, multiplication by is an injective -linear map because is a field. On the finite-dimensional space it is therefore surjective, so for some . Thus is a field containing and , hence contains ; conversely . Consequently and its dimension over is at most . If is finite, the tower law supplies the second bound.
Proof X-ray
How a span becomes a field
A span is only a vector space; two facts upgrade . Closure under multiplication comes from : a product of basis elements lies in and expands over , so -combinations multiply to -combinations. Inverses come from finiteness: multiplication by a nonzero is -linear and injective because has no zero divisors, and an injective linear self-map of a finite-dimensional space is surjective. Only enters; may have infinite degree over . Independence is never claimed, which is why the result is an inequality.
The proof also shows a practical method: reduce products of the chosen basis elements in before allowing coefficients from . Once spanning is known, one can decide whether the bound is sharp by testing independence over . One should not assume that a relation with coefficients in the larger field would already have coefficients in : the larger field has more scalars available.
Common mistake
(An upper bound is not automatically an equality)
The original -basis need not stay independent over . For example, if inside , then has degree one over , although . It would be wrong to multiply the two degrees automatically and announce .
Worked example 1.6.6
(Two quadratic fields and their compositum)
Take and inside . We first show that . If with , squaring gives . Since is irrational, . If then ; if then . Neither has a rational solution: in lowest terms the exponent of the prime in the square of a rational number is even, whereas it is odd in and in .
Therefore has no root in and is irreducible there. Thus and . A basis is
Indeed is a basis over , and is a basis for over ; multiplying the basis elements gives the displayed basis.
Set . Since , we have , and consequently and . Hence , so has degree four. From we get , or . The polynomial is monic of degree four and annihilates ; because has degree four, it is its minimal polynomial. The degree argument is what establishes irreducibility.
Common mistake
(The compositum depends on the chosen copies)
The degree is not determined by the isomorphism classes of and . Let be the real root and a primitive cube root of unity. The fields and are isomorphic over , both of degree three. The compositum of with itself is , of degree three. The compositum contains , which is nonreal and so lies outside ; it therefore equals , of degree over , strictly between and the bound of the theorem. Always say which copies inside which overfield are being combined.
Theorem 1.6.7
(Coprime degrees multiply)
Let and be extensions of inside a common field , with and finite. Then is a common multiple of and and is at most . In particular, if , then and .
If moreover , then , and is the degree of the minimal polynomial of over . This polynomial divides the minimal polynomial of over in .
Proof
(Two towers and one bound)
By the compositum bound, , so all the extensions below are finite. The tower gives , and the tower gives . Hence and both divide , and . When , every common multiple of and is a multiple of , and a positive multiple of that is at most equals . Dividing by gives .
For the second part, because contains and . Conversely, contains and , hence contains ; it also contains , so it contains . Thus , and is the degree of the minimal polynomial of over . The minimal polynomial has coefficients in and vanishes at , so divides it.
Worked example 1.6.8
(Irreducibility that survives an extension)
Take and inside . Here , and because is irreducible by Eisenstein's criterion at . The degrees are coprime, so .
Two irreducibility statements follow without any calculation with roots. Since and , the minimal polynomial of over is a monic divisor of of degree three, so it is : this polynomial stays irreducible over . Likewise shows that stays irreducible over .
This compositum is the field of the finite-witness example below, so its degree is exactly six, not merely at most six. The degree of the particular element is a separate question.
Coprimality cannot be dropped. In the preceding warning, and both have degree three and their compositum has degree six: a multiple of below the bound .
Theorem 1.6.9
(The algebraic elements form a field)
For any extension , the set of elements of algebraic over is a subfield of . Moreover, if an element of is algebraic over , it already belongs to ; one calls the relative algebraic closure of in .
Proof
(Use finite generation and then transitivity)
The set contains . If , the field is finite over , so , , and, when , are algebraic over . Thus is a field. If is algebraic over , transitivity applied to shows that is algebraic over , hence lies in . This does not assert that is algebraically closed: a polynomial over may have no root in .
Worked example 1.6.10
(A finite witness for one expression)
In a common algebraic closure of , choose positive real roots and , and let . The field has degree at most over , by adjoining the roots in that order. Thus the seven powers are linearly dependent over , and is algebraic of degree at most six. The bound does not by itself prove that its degree is exactly six.
Worked example 1.6.11
(An algebraic extension of infinite degree)
Let be the set of all complex numbers algebraic over . The preceding finite-generator argument applied to pairs shows that sums, products and inverses of nonzero elements of remain algebraic, so is a field. It is algebraic over by definition. For each , the polynomial is irreducible over by Eisenstein's criterion at , so contains an element of degree . If were finite, the tower law would bound all these degrees by , a contradiction.
Worked example 1.6.12
(An explicit infinite tower)
For let , choosing the positive real root. Since , these fields form an increasing chain, with . Eisenstein's criterion at shows that is irreducible over , so . The tower law then gives .
Their union is a field: any two of its elements already occur in one sufficiently large , where all field operations are available. Every element of the union is algebraic over , because it lies in some finite extension . Yet its degree over is infinite. A finite degree would bound for every , contradicting the unbounded sequence . This example exhibits finite witnesses concretely: the stage required depends on the element, and the degrees of the stages are not uniformly bounded.
Counterexample mode
Transitivity needs the lower extension to be algebraic
Claim: if is algebraic, then is algebraic for every subfield . Take , with transcendental over , and . Then is finite, even of degree one, and so algebraic. But is not algebraic over , so is not algebraic. The proof fails exactly at the step where the coefficients of an equation for over , here the coefficients of , were assumed algebraic over . The repaired statement requires to be algebraic as well.
Exercises with solutions
Quick check
(Suppose is algebraic over an algebraic extension . Explain why is finite, even when is infinite.)
Solution · Answer
Transitivity makes algebraic over , so it has a minimal polynomial over of finite degree, and that degree equals . The proof of transitivity uses only the finite extension of generated by the coefficients of one equation for , never all of .
Quick check
(Let be transcendental over and let be algebraic inside a common overfield containing . Why does remain transcendental over ?)
Solution · Answer
If were algebraic over , then and would both be algebraic, and transitivity would make algebraic over , contradicting the hypothesis. The common overfield is what gives meaning to both statements.
Quick check
(If consists of algebraic elements, must have finite degree?)
Solution · Answer
No. Take and , the field of all complex algebraic numbers from the example above. Then is algebraic but has infinite degree. Every particular element nevertheless lies in a finite subextension.
Quick check
(Let be finite with and let be any extension in a common overfield. What can be said about ? Must it equal five?)
Solution · Answer
The compositum theorem gives . Equality need not hold: taking gives and degree one. The upper bound counts a spanning family; it does not ensure that the family remains independent over .
Quick check
(Let and . Find and , and explain why is finite over .)
Solution · Answer
Direct multiplication gives , and . Both and lie in the degree-four field computed above, so their generated field is an intermediate field of finite degree by the tower law. In fact , so , the whole degree-four compositum.
Quick check
(In the quadratic compositum above, compute .)
Solution · Answer
The element is irrational: a rational square has even exponent at the prime , whereas has exponent one. Hence is its minimal polynomial and . Since lies in the larger field, which has degree four over , the tower law gives .
Quick check
(Show that is irreducible over , and find .)
Solution · Answer
The fields and have degrees two and three over ; the second degree comes from Eisenstein's criterion at . The degrees are coprime, so the compositum has degree six and has degree three over . This compositum is , so the minimal polynomial of over has degree three and divides . A monic divisor of the same degree is itself, which is therefore irreducible over .
Quick check
(Suppose and inside a common overfield. Which values of does the coprime-degree theorem allow? Give an example inside with the smaller value.)
Solution · Answer
The degree is a multiple of and at most , so it is or . For an example, take , of degree four by Eisenstein at , and , of degree six by the worked example above. Since , the compositum is , the compositum of fields of coprime degrees four and three. Its degree is .