Field Extensions and Finite Fields


What field is roots of in? They are obviously not in by Eisenstein's criterion or rational root theorem. And since we've already known are the solution of it, and it's in , so the roots are in . What a simple question.

But what if I ask, is there any smaller field than that contains ? Consider equipped with conventional addition and multiplication in : clearly it's closed under addition; since , it's closed under multiplication and thus a commutative ring; since , every non-zero element is invertible, and thus it's a field. It won't be hard to verify that is inside, so is a field that contains and smaller than .

What's the point of finding smaller fields such as rather than sleeping on ? You see, in order to reach the field , we cheat twice in definition. First, we cheat by allowing "infinite" times (defined by converging series) of addition, subtraction, multiplication and division to be closed in the real numbers , and then cheat by allowing even to be solvable by defining its root , and with some tricks of complex analysis, we've proved the fundamental theorem of algebra that all polynomials over can find all its roots lying in . This is just like when you take a taxi and the driver asks you the destination, you simply reply "the Earth", something that is definitely right (since you will need rocket for space travel) but hollow.

When I take out a random quintic equation, e.g. , by "not solvable by radicals" it doesn't mean there's no solution, it doesn't even mean there's no explicit formula to express the solution: clearly it has at least one root in , and we can approximate the root by leveraging the Newtonian method to a infinite series of summation, and the four roots remaining can be expressed by this root with radical solution of quartic equations. By "not solvable by radicals", I mean you cannot breaking down into some radical equations (in the form of ), while all quadratic, cubic and quartic equations can. To judge whether an equation is convertible into radical equations, especially for quintic equations and higher, we need Galois theory, in which finding and studying "smaller fields than " is the baby step, but a giant step.

Field Extensions

A field extension is a binary relationship between base field and extension field such that is a subring (or we can safely use subfield in this scenario) of . The slash symbol can't be intepreted as quotient field of over .

Conventionally, We also adapt the symbol to represent the relationship of two or more fields chained together by field extensions (aka. Tower of fields), where the subset symbol does not merely mean by set inclusion.

We can come up with some instances of field extensions quite quickly: like , and . At the first glance you may be wondering what good will such a simple relationship do, but we are about to add some important details to it.

Smallest subfields of different characteristics

We would like to prove that any field of characteristic is an extension field of (which is not necessarily a proper subfield), such that is isomorphic to .

First, given that is of characteristic , we have a cyclic subgroup generated by , and obviously there's . Since every element in can be written as , by equipping it with ring multiplication, following the ring's axioms there must be , so is closed under multiplication and isomorphic to .

Then, obviously is integral domain, since we've proved that any field containing integral domain must also include a subring isomorphic to 's fields of fractions , we have , and we are done.

With analogous rationale, we can prove that any field of characteristic is an extension field of such that 1: the cyclic subgroup generated by is , by equipping it with ring multiplication we have , finally we have .

In this way, the subfield related to and created using the methods we've described above is the smallest. The term "smallest" means you can't make smaller subfield, as a field must always contain the unity , with the addition of made closed while the elements inside made invertible. And the structure of such smallest subfield is completely defined by the characteristic of field: a field of characteristic contains the smallest subfield isomorphic to ; a field of characteristic contains the smallest subfield isomorphic to .

On the other hand, since the process of field extension will never mutate the ring operations defined in the smallest subfield, a field extension should never exists if .

Simple extensions

The process of human's finding solutions to algebraic equations is the process of finding "new numbers". Through solving the equation by the Pytagorean theorem, Hippasus discovered the root is a "new number" that is not in . Today we know it's an irrational number and in or . Through radical solution to cubic equations, Cardano adopted the existence of square root of , which is also a "new number" to . Although Cardano said it was "an agony" dealing with the "new number" and took only root in real numbers, people were unaware of the usefulness of the "new number" until the proof of fundamental theorem of algebra by Gauss, naming the "new number" to imaginary unit and enlarging the number field to .

The phenomenon behind finding "new numbers" is astonishing: whenever there's polynomial with no root in the base field, an extension containing the root field always come up in time, e.g. has root , has root .

While it is not just a coincidence, we will present the theorem that for a field and an irreducible polynomial such that , there's always a field extension such that .

Given that there're already real instances of this theorem, e.g. and ), why don't we observe them first? In the field , we can always use as a term that remains intact under the field operations of , except for replacing with . Analogous thing happens for , in which we will need to replace with . Actually, it won't be hard to find that we can operate and in the way just like the indeterminate of the polynomial ring . When it comes to replacement, we will just need to the evaluate the canonical homomorphisms of and , where there're and . Finally, it won't be hard to verify that defined by and is field isomorphism, and so does defined by and . In this way, and describe the structure of and completely.

Such observation can be elevated to our case: since is constrained to be irreducible polynomial, by the theory of polynomial rings we know is maximal, therefore is field. By our assumption, if the extension field truly exists, it won't be hard to verify that there's ring homomorphism defined by and . Since is already a field, the kernel of is only zero or the whole field, while it's clearly not the latter case, so must contain a subfield isomorphic to . In the meantime, is the root of , which can be easily verified by:

In this way, the field is the extension field of such that , so the existence of the extension field is proved. However, we would like to take one more step to claim and prove that , which means it's an -dimensional vector space over . From the theory of polynomial rings we know , where all polynomials , are mapped to distinct coset , and all polynomials are mapped to with being obtained by dividing by , using the division algorithm of . And by applying the ring isomorphism of we have .

The field is the extension field 2 of we want, and is called a simple extension of , obtained by adjoining element to . Clearly the polynomials associate with shares the same root of and generate the same principal ideal as , for convenience we pickup the monic one among them and call it the minimal polynomial of over , while is called the degree of over , describing the dimension of when it's viewed as finite dimensional vector space over .

One can easily verify that . So for covenience, we allow to appear in the slot of element to adjoin such that , where the minimal polynomial of over is , and the degree of over is .

Consider a simple extension where the minimal polynomial of over is , we would like to prove any polynomial that has root in must be divisible by . Let's do Euclidean algorithm in so that we have . Given that is irreducible, might only be or . Assume it's the former case, then we have , which is a contradiction. So it's only the latter case and we are done.

In this way, we can prove the minimal polynomial of over is unique: assume there were distinct minimal polynomials of over , we have ; since both and are irreducible, they can only be associate; since they are both monic, there can only be , which is a contradiction to 's distinction.

Conversely, the element defined by minimal polynomial may not be unique: both are the root of ; both are the root of ; and both are the root of . More interestingly, in the field extension that contains distinct roots of the same minimal polynomial , given that there's , the extension fields obtained by adjoining to must be isomorphic, and with little effort one will see the isomorphism is defined by and . Such pair of are said to be conjugate over . This means one may not tell one extension field from another without prior knowledge of and operating merely on "alphabetic" level (that is, presuming the root of to be in ).

Algebraic extensions and transcendental extensions

In a field extension , an element is said to be algebraic over if it's the solution to any algebraic equation in ; otherwise it's said to be transcendental. The field extension is said to be algebraic extension if all elements in is algebraic over , and transcendental extension otherwise.

We can quickly come up with some instances: the extension is algebraic, by every equation can be converted into , and ; the extension is transcendental, since is transcendental 3 and not the root of any algebraic equation in . An transcendental extension may contain algebraic elements (e.g. ), but must contain at least one transcendental element.

For a field extension that with an element , then let be the smallest field including every elements in and the element , we would like to show that is algebraic iff there's where is the minimal polynomial of over , and is transcendental iff there's .

To make the smallest field, we first take the "molds" of possible addition, subtraction and multiplication operations for the element to adjoin with existing elements in , and the "molds" form the set ; then we put the element into the molds to evaluate the smallest ring, this is done by evaluation homomorphism with its domain trimmed to , resulting in the ring ; finally any field that contains must also contain its field of fractions, so the smallest field containing and must be isomorphic to .

When is algebraic, we can find its minimal polynomial over ; and since we know when 's domain is trimmed to , we can draw the conclusion that ; finally since is already a field, we have , and therefore the simple extension of adjoining coincides with the smallest field containing and .

When is transcendental, there's no non-zero polynomial such that ; so we know there's , and thus ; finally by we can draw the conclusion that the smallest field containing and is isomorphic to , which is the field of rational functions over .

Finally, since 's being algebraic or transcendental are mutually exclusive and complementary, there's no other case of structure of and we can safely identify whether is algebraic or transcendental by the sturcture of .

Finitely generated field extensions

In the field extension , for elements , we define as the smallest field containing and , and call it the finitely generated field extension. Obviously such a smallest field can be formed by substituting in indeterminates of with . But we would like to show such a smallest field can also be formed by extending with consequently in arbitrary order, by adjoining elements one at a time.

First we would like to show the equivalence of indeterminates for substition, that there's , assume they are not equal by some rational function such that , but given that there's rational function (tips: define a rational function that takes in the parameter in the order of mitigating the permutation by ) such that , the assumption is false.

Then we would like to show the field can be formed by adjoining to , that there's . First we have by simply aggregating coefficients of , thus we have under the meaning of set inclusion. Conversely, we have by the operational rules of , thus we have under the meaning of set inclusion. Finally, by they contain each other under set inclusions, we have .

In this way, we've shown that can be obtained by adjoining to in arbitrary order, forming a tower of fields that .

Finite extensions

Let's consider an instance of finitely generated field extension, that on the top of , we may adjoin to it, forming the field which is also also a finite dimensional vector space over the base field with basis , which is obtained by multiplying basis of over and the basis of over . The same thing will happen if we adjoin to , as we've proved.

In order to generalize such kind of field extensions, we introduce the concept of finite extension: field extension is said to be a finite extension when is finite dimensional vector space over , with its dimension onventionally denoted as .

We would like to prove that all finite extensions are algebraic. Let there be , assume there's transcendental element , then for any , there's by 's transcendence, which means forms a basis of -dimensional vector space. This is not true when , since is -dimensional vector space over , and the elements taken from are always linearly dependent, impossible to build up the basis.

And we would like to prove the Tower law, that if are both finite extensions, is also finite extension with . Let there be , the basis of be , so that the elements of are in the form of , and the basis of be , so that the elements of are in the form of . So eventually the elements of can be represented as .

To prove that are linearly independent, assume they are linear dependent instead by for a set of that're not all zeroes. Combine the coefficients of s by , since are linearly independent, not all of the are zeroes. But this will imply for a set of that are not all zeroes, and that are linearly depdent, which is a contradiction. In this way, we've proved that is a finite extension of with basis , and of dimension .

The instance we've came up with is a also a good instance of the Tower Law, that there's while and .

Finally, we would like to show that every finitely generated algebraic extension, the finitely generated field extension with all elements to adjoin being algebraic, is a finite extension, and every finite extension is a finitely generated algebraic extension. Consider the finitely generated algebraic extension , the intermediate extensions are simple extension and thus finite extensions, and by the Tower Law we know must also be a finite extension. Conversely, let there be finite extension with basis , and all elements of are in the form of , which must be in . But if we allow to be proper subset of , given that are all elements of and their field operations are made closed in , it's a contradiction to 's being smallest field containing and , so there must only be . And since are all algebraic, must be a finitely generated algebraic extension. In this way we've shown the interconnection between finitely generated algebraic extensions and finite extensions.

Splitting fields

Inspired by the fundamental theorem of field extensions, for any algebraic equation , which can be asserted to be monic, without losing generality, we try to extract the roots of by factorizing into over some "sufficiently large" field, using the following algorithm:

  1. Initialize .
  2. If , halt; otherwise increment by .
  3. Take out any irreducible factor , of over .
  4. If is of degree , then there's , so we just need to update directly and go back to step 2.
  5. Otherwise has no root in , but we can make simple extension such that is the minimal polynomial of over . So we will update and go back to step 2.

In the algorithm, we factorize by plucking factor off , one at a time, resulting in a factorization of with exactly factors of over that's "sufficiently large". Thus the algorithm will halt exactly at with .

Apparently the algorithm sheds light on the existence of the splitting field of , which is the smallest field (not just "sufficiently large", but perfect at size) containing all roots of . We can also say that splits .

Problem of uniqueness of splitting field

Let's evaluate the splitting field of following the steps of our algorithm. It's already irreducible by Eisenstein's criterion, and is one of the roots and on the number axis. But given that there're other conjugate roots off the number axis on complex plane, the extension field does not split . But now we have over . The roots of are the other two conjugate roots of , where is the primitive cubic root. And by merely adjoining to we get , in which there's . So is the splitting field of with .

On the other hand, you can first adjoin to , and then there's , where the contains the other two conjugate roots and is irreducible for being a cyclotomic polynomial of prime order. And when we adjoin to , clearly the contains all of the roots and thus splits . In fact, given that we've known the three roots , simply adjoin all of them to and then can be expected.

In general, given a field extension where contains all roots of any polynomial (the is referred as the algebraic closure of , which we will not formalize in this text immediately though), e.g. , then the splitting field of must be uniquely . Otherwise for any other splitting field of whose roots differ from for at least one element. So has roots, or is not UFD by , choose your favorite contradiction. In such manner, given that splitting field of must be unique as with 's roots in being , what's problem of "uniqueness of splitting field", as is mentioned in the title?

The problem is that we presume there's a unique algebraic closure of , identified by set elements. Such hallucination is rooted in our stereotype that polynomials over should be eventually splitted over . To cure the hallucination, we need our old friend .

Consider the field and algebraic equation over . Clearly the root is not in and we need to extend. For the extension field, apparently a candidate would be , which splits for containing both . Meantime another candidate is , where every non-zero element is invertible by . It also splits for containing both roots . There will never be a unique algebraic closure containing both and at the same time despite their being "conjugate", otherwise is observed. Therefore neither the uniqueness of splitting field nor the uniqueness of algebraic closure could be expected. The uniqueness is identified by set elements.

By now, should we conclude that the splitting field is not unique? Well, it is still in the midway of the story. The two splitting fields are actually isomorphic by , as their structure are totally defined by the minimal polynomial . In fact, we are able to prove the diverging splitting fields are isomorphic, and their uniqueness can be identified by field structure.

Splitting field is unique up to isomorphism

It won't be hard to verify that there's , while the "roots" are defined by , and the minimal polynomial and of them are interchangeable under the isomorphism of . In fact, such observation is the starting point of the whole proof.

Consider there're two isomorphic fields connected by field isomorphism , it won't be hard to see there's an induced field isomorphism defined by . Obviously for an irreducible polynomial , must also be irreducible, and vice versa. So both and must be fields. Given that there's , by observing the cosets we can simply conclude that . Finally by simply substituting in the roots defined by and , we can conclude that .

For polynomial , the existence of some field that's large enough to contain all roots of is obvious. Without losing generality, assume there're distinct fields that contain all roots of , we can pluck off , which are both defined by the same minimal polynomial over , there must be , defining field isomorphism such that . And for any factorization over , there must also be over where is the field isomorphism defined by . If already splits , then we are done. Otherwise we can pluck off , which is defined by minimal polynomial , meantime we pluck off defined by minimal polynomial . In this way, there must be defining field isomorphism , and defining field isomorphism . Continue on such process of factorization until splits in , there must be an isomorphic field splitting in , while internally is unique in and is unique in . In this way, we've proved that the splitting field of must be unique up to isomorphism.

And clearly the proof does not rely on identical base fields, for isomorphic base fields connected by field isomorphism , the statement still holds that for the splitting field of and splitting field of , there's , where is induced ring isomorphism between polynomial rings by . This establish the statement that splitting field or of in our example and the splitting field of are the same up to isomorphism.

Finite fields

One of the most important object in field theory is finite field. A finite field is, namely a field of finite order. We may also call it the Galois field in memory of Evariste Galois. It can be shown that a finite field can only be of order , where is prime and is positive integer, and all finite field of the same order are isomorphic. Such non-triviality makes finite field prominent.

Finite field can only be of prime power order

For a finite field , since it's finite, it cannot be of characteristic , so its characteristic must be some prime , and we've already shown it contains a subfield .

Let and be elements outside . Clearly there's no transcendental element in over , so we can make finite extension , which is of order when . On the other hand, all elements of has been adjoined, so it can only be 4. In this way, when is of characteristic , and , the order of can only be .

Separable polynomials

A polynomial is said to be separable when it has no duplicate root in its splitting field. Such kind of polynomials plays an important role in proving the isomorphism of finite field of the same order.

The formal derivative on polynomial ring is a map defined by , We would like to claim and prove that polynomial is separable iff .

Let's derive some basic properties of the formal derivative that we will use first:

So the formal derivative and the derivative in calculus have operation rules in common. However formal derivative does not rely on the feasibility of defining limits, which is generally not possible for fields, especially for finite fields.

Let separable polynomial be factorized into with distinct roots in its splitting field. It won't be hard to find there's:

And for every irreducible factor of , it divides every factor except for . So whenever is separable, there's .

Conversely, without losing generality, let polynomial be of duplicate root , and it won't be hard to find there's:

Apparently there's , and we are done.

Finite fields of the same order are isomorphic

More precisely, we would like to show all fields of order are isomorphic to the splitting field of .

First, it won't be hard to find there's , and thus , which means the polynomial must be separable in its splitting field, with distinct roots.

Then we would like to show that the roots of forms a field. One should remember the Freshman's dream that in commutative ring of characteristic . And given that , we can repeat such process so that there's . In this way, we can easily conclude there's in of characteristic . Assume are the roots of , we have:

  1. and .
  2. .
  3. .
  4. When , .
  5. When , .
  6. When , .

In this way, the roots of forms a subfield of its splitting field, with elements and it's the extension field of . However, recall how will we make a splitting field: for roots outside , we adjoin it to until it splits , and we've already known the elements in the splitting field are obtained by field operations of with roots outside . We have just proved they will result in another root, thus the splitting field is subfield of , and is the splitting field of itself. In this way, we've proved the existence of the finite field of order , whose structure is the splitting field of polynomial .

For the uniqueness of finite field of order , first we know its multiplicative group is cyclic, so every elementsin fulfils , and every element in fulfils including . Then we also know contains a subfield , and can be viewed as a polynomial over . When we "grow" the splitting field of towards , by adjoining root of current irreducible polynomials merely from until it splits , we will find the splitting field is a subfield of with the same amount of elements as , so coincides with the splitting field of . Finally, since and both and are splitting fields of the same polynomial up to isomorphism, we have .

In this way, we've proved the finite field of the same order must be unique up to isomorphism, thanks to the uniqueness of splitting fields. Conventionally, we denote the unique finite field of order as or . Obviously we can derive from the process of deduction above.

Subfields of finite fields

We would like to show contains a unique subfield isomorphic to iff .

First, if would like to contain any subfield, clearly it can only be , and both and are finite extensions. By the Tower law, we have , so such a subfield exists only if there's .

Then, for the existence of such subfield, consider the modulo- reduction of polynomial to , obviously the image of factor polynomial of over is the factor polynomial of over . And we know there's over , then we have over . Obviously is also separable so there're roots contained in the factor of . Obviously those elements in are roots of , we can "grow" the splitting field in the direction of splitting first, and then clearly the subfield that splits is exactly the same as . In this way, has a subfield isomorphic to if there's .

Finally, for the uniqueness of such subfield, obviously the is the subgroup of . Since the multiplicative group is cyclic, the subgroup of order is unique as , so the subfield must also be unique.

Finite fields can be obtained from simple extensions

We would like to show that for every finite field , let be the primitive root and be the minimal polynomial of over , then . Which means every finite field is extension field obtained by a simple extension adjoining primitive root over .

First, obviously is algebraic over , by the theory of algebraic extension we know there's , all we need to do is to find the image of .

Then, for the evaluation homomorphism , from the domain of coefficient and the element taken into evaluation we know the image is subset of . And by and , we know the is surjective on its image , thus we have and .

Finally, when these two pieces are putting together, we can conclude that there's and we are done.

The importance of this theorem is, it provides a very simple way to visualize finite fields of and do field operations on it, which is done by polynomial ring operations of modulo minimal polynomial of the primitive root.

Conclusion

In this text, we discuss some basic properties of field extensions. First enlightened by human's solving equations we discovered a kind of field extension called the simple extension, claiming that for each irreducible polynomial that has not root over the base field , there's inevitably a extension field contains a root of , by the PID property of polynomial ring over . Then we introduced the concept of algebraic and transcendental elements and extensions, showing the structure of field with single algebraic or transcendental adjoined to it. Finally we introduced the concept of finitely generated extensions and finite extensions, proving that finitely generated extension adjoining the same set of elements to base field is unique and all finite extension must be finitely generated, and vice versa.

We then dig into human's solving equation and introduced the concept of splitting field, which is the smallest field containing the base field and all roots of the equation, and splits the polynomial of the equation. We also show that the splitting field is always not unique when identified by set elements, but always unique up to isomorphism.

Finally, equipped with tools of field extensions we explore the properties of finite fields. We've shown that all finite fields has the order of prime power , and isomorphic to the splitting field of . We've shown finite field has a unique subfield iff there's . And we've shown every finite field can actually be obtained by simple extension of adjoining primitive root , providing an alternative way of doing field operation as doing polynomial ring operations modulo minimal polynomial of the primitive root.

The concepts and properties of field extensions and finite fields are the foundation of fields and Galois theory.


  1. The symbol is introduced in the proof of existence theorem of primitive roots in finite fields, which is also a crucial conclusion for fields. [return]
  2. Some reader might be confused about when and why to use the symbol rather . Well, in this specific scenario, it's in the first place; and since it's a field, the field of fraction of it is isomorphic to it, or we can say . More generally, we will use whenever it's a field, and whenever it's not a field. Like and are not fields, and we should never use round bracket in these scenarios. [return]
  3. I won't prove the transcendence of in this text, since it's too complex and not proving it will not hurt the understanding. For more details, see also the Lindemann–Weierstrass theorem. [return]
  4. Alternatively, one can also apply the method we used in proving finite division rings are vector spaces over their centers. Starting from the subfield , we select and expand the vector space until it covers the whole finite field. [return]
October 22, 2023