The Field of Fractions

Can you always divide? In the integers you cannot solve $2x = 3$ . The rational numbers fix this by formally adjoining fractions $a/b$ . This article asks a question. Given any integral domain, can we perform the same construction? In what sense is the result the smallest field containing it? The idea matters because it is the archetype of a "free" construction in algebra. We start from a modest algebraic object with just enough structure (no zero divisors). From this we build the most economical field extension possible. We do so functorially. This means any map out of the domain into a field automatically extends. This same pattern reappears constantly. Build an equivalence class of formal expressions, define operations, then verify a universal property. This happens most notably when we later build the field of rational functions $k(x)$ over a polynomial ring $k[x]$ . The big picture is that "integral domain" is exactly the right hypothesis. It makes this construction work. It is precisely the condition that makes the naive fraction-building recipe (cross-multiplication) into a genuine equivalence relation. That relation is compatible with arithmetic. It is also precisely the condition needed to embed the domain into the resulting field. Losing it (allowing zero divisors) causes the construction to collapse or misbehave. We verify this explicitly. This article delivers a complete construction of the field of fractions $\operatorname{Frac}(R)$ of an integral domain $R$ . It gives a full proof that it is a field containing $R$ . It also gives a rigorous statement and proof of its universal property. $\operatorname{Frac}(R)$ is the unique smallest field in which $R$ embeds. Every injective-on-domain homomorphism from $R$ into a field factors uniquely through it.

Key takeaways

Given an integral domain $R$ , we formally invert all nonzero elements by building equivalence classes of pairs $(a,b)$ , $b \neq 0$ , thought of as $a/b$ .
The equivalence relation $(a,b) \sim (c,d) \iff ad = bc$ requires $R$ to have no zero divisors to be transitive and well-defined with the arithmetic.
The resulting set $\operatorname{Frac}(R)$ with the usual fraction addition and multiplication is a field, and $R$ embeds into it via $r \mapsto r/1$ .
$\operatorname{Frac}(R)$ satisfies a universal property: any injective ring homomorphism from $R$ into a field $K$ extends uniquely to a field homomorphism $\operatorname{Frac}(R) \to K$ .
This universal property characterizes $\operatorname{Frac}(R)$ up to unique isomorphism fixing $R$ , so the construction, though built from arbitrary representatives, produces a canonical object.
Key idea: integral domains are exactly the rings that embed in a field, and $\operatorname{Frac}(R)$ is the universal such field.

Where this lives today

The field of fractions is the simplest instance of localization, a fundamental tool throughout commutative algebra and algebraic geometry: instead of inverting all nonzero elements, one inverts a chosen multiplicative set $T \subseteq R$ to get $T^{-1}R$ , with the same fraction-and-equivalence-class recipe (modified to handle possible zero divisors in $T$ ). Localization underlies the construction of local rings at a point of a variety, the study of primary decomposition, and the definition of the structure sheaf in scheme theory. In polynomial rings over a field, $\operatorname{Frac}(k[x_1,\dots,x_n])$ is the field of rational functions $k(x_1,\dots,x_n)$ , indispensable in algebraic geometry (function fields of varieties), in the theory of transcendence degree, and in the definition of rational maps between varieties. In number theory, $\operatorname{Frac}(\mathcal{O}_K)$ for a ring of integers $\mathcal{O}_K$ recovers the number field $K$ itself, and the construction underlies the definition of $p$ -adic fields as completions built from fraction-field-like limiting procedures. In cryptography, arithmetic in finite fields and their extensions (built via polynomial ring quotients) sometimes requires passing to fraction fields of polynomial rings over finite fields, e.g., in coding theory and the study of rational function fields $\mathbb{F}_q(x)$ used in constructing algebraic-geometry codes. In computer algebra systems, exact rational arithmetic and symbolic manipulation of rational expressions are direct computational implementations of $\operatorname{Frac}(\mathbb{Z})$ and $\operatorname{Frac}(k[x])$ .

:::faq[How does $\operatorname{Frac}(k[x])$ relate to rational functions we know from calculus?] $\operatorname{Frac}(k[x])$ , denoted $k(x)$ , is the algebraic field of formal ratios of polynomials, whose elements are equivalence classes under cross-multiplication rather than functions. When $k = \mathbb{R}$ or $\mathbb{C}$ , one can evaluate a representative $f/g$ at any point where $g$ does not vanish, and this evaluation is consistent across equivalent representatives wherever both denominators are simultaneously nonzero — this is why the calculus notion of "rational function" and the algebraic element of $k(x)$ agree in practice, though the algebraic object exists independent of any notion of evaluation and is cleaner for purely algebraic manipulations like field extensions and Galois theory.

Ask the tutor: How does this construction generalize to localizing a commutative ring at an arbitrary multiplicative set, and what changes when that ring has zero divisors?

R

Intuition

The construction mirrors exactly how a child learns to build $\mathbb{Q}$ from $\mathbb{Z}$ : a fraction $a/b$ is not really a pair of integers, it is an equivalence class of pairs, since $1/2$ , $2/4$ , and $-3/(-6)$ should all denote "the same number." The rule $ad = bc$ is just cross-multiplication, the familiar test for whether two fractions are equal. What integral domains buy us is that this cross-multiplication test behaves like an honest equality: it is reflexive and symmetric for free, but transitivity — the fact that $a/b = c/d$ and $c/d = e/f$ should force $a/b = e/f$ — genuinely uses the absence of zero divisors, as the proof below shows. Once the set of fractions is built, addition and multiplication are defined exactly as with rational numbers: find a common denominator, multiply numerators and denominators. The nontrivial content is that these operations are well-defined — independent of which representative pair we chose — and that they endow $\operatorname{Frac}(R)$ with the structure of a field, with $0/1$ and $1/1$ as additive and multiplicative identities and $b/a$ as the inverse of a nonzero $a/b$ . The picture below visualizes $\operatorname{Frac}(\mathbb{Z}) = \mathbb{Q}$ : each rational number is an equivalence class of lattice points $(a,b)$ with $b \ne 0$ , all lying on a common ray through the origin (since $(a,b) \sim (c,d) \iff ad = bc$ means the points are collinear with the origin, i.e. $a/b = c/d$ as slopes).

Figure

Equivalence classes of pairs as rays through the origin: (1,2), (2,4), (3,6) all represent the fraction 1/2

Central results

R

\sim

R

\Phi

Worked examples

Example 1: The field of fractions of $\mathbb{Z}[i]$ , the Gaussian integers. Let $R = \mathbb{Z}[i] = \{a+bi : a,b\in\mathbb{Z}\} \subset \mathbb{C}$ . This is an integral domain (it is a subring of the field $\mathbb{C}$ , and subrings of a field with $1\neq0$ have no zero divisors, since a zero divisor in $R$ would witness $xy=0$ with $x,y \neq 0$ in $\mathbb{C}$ too, impossible in a field). By the universal property, take $K = \mathbb{C}$ and $\varphi: \mathbb{Z}[i] \hookrightarrow \mathbb{C}$ the inclusion, which is injective. This extends uniquely to $\Phi: \operatorname{Frac}(\mathbb{Z}[i]) \to \mathbb{C}$ via $\Phi(a/b) = a \cdot b^{-1}$ computed in $\mathbb{C}$ . We claim the image of $\Phi$ is exactly $\mathbb{Q}(i) := \{p+qi : p,q \in \mathbb{Q}\}$ , and that $\Phi$ gives an isomorphism $\operatorname{Frac}(\mathbb{Z}[i]) \cong \mathbb{Q}(i)$ . Image is contained in $\mathbb{Q}(i)$ : take a representative $a/b$ with $a = a_1+a_2i,\ b=b_1+b_2 i \in \mathbb{Z}[i]$ , $b \neq 0$ . Rationalize:

\Phi(a/b) = \frac{a}{b} = \frac{a\bar b}{b\bar b} = \frac{(a_1+a_2i)(b_1-b_2i)}{b_1^2+b_2^2} = \frac{a_1b_1+a_2b_2}{b_1^2+b_2^2} + \frac{a_2b_1 - a_1b_2}{b_1^2+b_2^2}\,i,

and since $b \neq 0$ we have $b_1^2+b_2^2 \neq 0$ (a sum of two integer squares, not both zero), so this exhibits $\Phi(a/b) \in \mathbb{Q}(i)$ explicitly. Concretely, take $a = 1+2i$ , $b = 3-i$ (so $b_1=3, b_2=-1$ , $b_1^2+b_2^2 = 10$ ). Then

\Phi\!\left(\frac{1+2i}{3-i}\right) = \frac{(1+2i)(3+i)}{10} = \frac{3+i+6i+2i^2}{10} = \frac{3+7i-2}{10} = \frac{1+7i}{10} = \frac{1}{10}+\frac{7}{10}i.

Image contains all of $\mathbb{Q}(i)$ : given $p+qi$ with $p=m/n, q=k/n$ (common denominator $n \in \mathbb{Z}_{>0}$ , $m,k\in\mathbb{Z}$ ), we have $p+qi = (m+ki)/n = \Phi\big((m+ki)/n\big)$ where $m+ki, n \in \mathbb{Z}[i]$ , $n \neq 0$ . So $\Phi$ is onto $\mathbb{Q}(i)$ . Since $\Phi$ is injective (by the theorem) and surjective onto $\mathbb{Q}(i)$ , we conclude $\operatorname{Frac}(\mathbb{Z}[i]) \cong \mathbb{Q}(i)$ .

Example 2: The field of fractions of $R = k[x]$ , giving $k(x)$ , evaluated at an explicit non-trivial fraction. Let $k$ be a field (say $k=\mathbb{Q}$ ) and $R = k[x]$ , the polynomial ring in one variable, which is an integral domain because $k$ is a field (the leading coefficients of a product of nonzero polynomials multiply to a nonzero element of $k$ , by no-zero-divisors in $k$ , so the product is nonzero). Its field of fractions is denoted

k(x) = \operatorname{Frac}(k[x]) = \left\{ \frac{f(x)}{g(x)} : f,g \in k[x],\ g \neq 0 \right\}\Big/\sim,

the field of rational functions. Take the explicit fraction

\frac{f}{g} = \frac{x^3 - 2x + 1}{x^2 - 1} \in \mathbb{Q}(x).

Let's verify this is not the zero element and simplify it using the equivalence relation. First factor: $g = x^2-1=(x-1)(x+1)$ , and $f = x^3-2x+1$ . Testing $x=1$ : $f(1) = 1-2+1=0$ , so $(x-1) \mid f$ . Polynomial division gives

x^3-2x+1 = (x-1)(x^2+x-1).

\frac{f}{g} = \frac{(x-1)(x^2+x-1)}{(x-1)(x+1)}.

We claim this equals $\dfrac{x^2+x-1}{x+1}$ in $\operatorname{Frac}(k[x])$ : by definition of $\sim$ , we need

(x-1)(x^2+x-1)\cdot(x+1) = (x-1)(x+1)\cdot (x^2+x-1),

which holds trivially by commutativity of $k[x]$ — this is the algebraic incarnation of "cancel the common factor $(x-1)$ ," valid here precisely because $k[x]$ is an integral domain, so the cross-multiplication test is a genuine equivalence, and cancellation of a nonzero common factor from numerator and denominator never changes the class. This also illustrates that elements of $\operatorname{Frac}(k[x])$ are not literally functions on all of $k$ : the original expression $f/g$ is undefined at $x=1,-1$ (denominator zero), while the simplified form $\frac{x^2+x-1}{x+1}$ is defined at $x=1$ (value $1$ ), yet both represent the same element of the field $k(x)$ — equality in $\operatorname{Frac}(R)$ is an algebraic identity of formal fractions, not pointwise equality of evaluated functions, and one must fix this distinction carefully when later interpreting elements of $k(x)$ as functions on a suitable domain.

R = \mathbb{Z}/6\mathbb{Z}

Scope, edge cases & common traps

Situation	What happens
$R$ is already a field	$\operatorname{Frac}(R) \cong R$ via $\iota$ , since every nonzero $b\in R$ already has $b^{-1}\in R$ , so $a/b = ab^{-1}/1 = \iota(ab^{-1})$ ; $\iota$ is onto.
$R = \mathbb{Z}$	$\operatorname{Frac}(\mathbb{Z}) = \mathbb{Q}$ , the motivating and defining example.
$R$ has zero divisors	Construction fails at the equivalence-relation step (see counterexample); no field of fractions exists in general.
$R$ is an infinite integral domain that is not Noetherian, not a UFD, etc.	Construction still works verbatim: the theorem uses only "integral domain," no finiteness or factorization hypotheses.
Confusing $\operatorname{Frac}(R)$ -equality with function equality	For $R=k[x]$ , $f/g = f'/g'$ in $k(x)$ need not mean $f(a)/g(a) = f'(a)/g'(a)$ pointwise wherever both sides are defined and nonsingular — though it does agree wherever both are defined, by the cross-multiplication identity being a polynomial identity.
Applying the universal property with a non-injective $\varphi: R \to K$	The theorem's hypothesis " $\varphi$ injective" is essential (see warning below).

Common trapThe universal property requires

\varphi

injective — not merely a homomorphism

A common error is to apply the universal property to any ring homomorphism $\varphi: R \to K$ into a field $K$ , forgetting the injectivity hypothesis. If $\varphi$ is not injective, the formula $\Phi(a/b) = \varphi(a)\varphi(b)^{-1}$ can fail to be well-defined or $\varphi(b)$ may be $0_K$ for some nonzero $b$ , making $\varphi(b)^{-1}$ meaningless. For instance, the reduction map $\mathbb{Z} \to \mathbb{Z}/2\mathbb{Z} = \mathbb{F}_2$ is a ring homomorphism into a field, but it is not injective (kernel $= 2\mathbb{Z}$ ), and indeed there is no ring homomorphism $\mathbb{Q} \to \mathbb{F}_2$ at all extending it: any such map would have to send $1/2 \mapsto \varphi(2)^{-1}$ , but $\varphi(2) = \varphi(1)+\varphi(1) = 1+1 = 0$ in $\mathbb{F}_2$ , which is not invertible. The correct general statement, that injective homomorphisms into fields extend uniquely, is precisely what is proved above — always check injectivity before invoking it.

The mathematician behind it

Origins · 1 minRichard Dedekind and how this was discovered

Richard Dedekind

1831-1916 · late 19th-century Germany, foundations of algebraic number theory

Dedekind did not "invent" fractions, but he was among the first to treat the passage from a domain of integers to its field of quotients as a general, rigorous algebraic construction rather than a numerical convenience. Working in Göttingen in the circle of Gauss, Dirichlet, and Riemann, he developed the modern theory of ideals in rings of algebraic integers, and the systematic construction of quotient fields — generalizing $\mathbb{Z} \to \mathbb{Q}$ to arbitrary rings of integers in number fields — was a key technical tool in that program, later abstracted fully in the language of commutative rings by Emmy Noether's school.

History

- 1858–1871: Dedekind develops the theory of ideals for rings of integers in algebraic number fields, implicitly using quotient-field constructions to define fractional ideals.

- 1893: Dedekind's treatment of algebraic number theory (in his supplements to Dirichlet's lectures) crystallizes ring-theoretic language later formalized abstractly.

- 1920s: Emmy Noether and her school (including Krull) formalize the abstract notion of a commutative ring and prove the field-of-fractions construction for general integral domains in the modern axiomatic style.

- 1930: B.L. van der Waerden's Moderne Algebra presents the construction of

\operatorname{Frac}(R)

in the textbook form essentially used today, cementing it as standard curriculum.

- Present: the construction is the prototypical example of a "localization" of a ring, vastly generalized in commutative algebra to localizing at arbitrary multiplicative sets.

Why must

R

be an integral domain, not just a commutative ring, for this construction?

The equivalence relation $(a,b)\sim(c,d) \iff ad=bc$ needs transitivity to define genuine equivalence classes, and the proof of transitivity requires cancelling a common nonzero factor: from $d(af-be)=0$ with $d \neq 0$ , we need to conclude $af=be$ . This cancellation is valid exactly when $R$ has no zero divisors. If $R$ has zero divisors, as shown for $\mathbb{Z}/6\mathbb{Z}$ , transitivity can fail outright, so $\sim$ is not even an equivalence relation and $\operatorname{Frac}(R)$ cannot be defined by this recipe. More conceptually, any ring embedding into a field must itself have no zero divisors (since fields have none and injectivity reflects this property), so integral domains are exactly the rings capable of embedding in a field at all.

\operatorname{Frac}(R)

literally unique, or just unique "up to isomorphism"?

It is unique up to a unique isomorphism that is the identity on $R$ , which is the strongest reasonable sense of uniqueness for a universal construction. Concretely, if $K$ is any field containing $R$ as a subring such that every element of $K$ is a ratio of two elements of $R$ , the universal property theorem produces a unique field isomorphism $\operatorname{Frac}(R) \to K$ restricting to the identity on $R$ . So while the explicit construction (equivalence classes of pairs) is one specific model, any other model satisfying the same universal property is canonically identified with it — this is the standard "universal property pins down the object uniquely" phenomenon seen also for quotient groups, tensor products, and free groups.

Does

\operatorname{Frac}(R)

depend on which representative pair

(a,b)

we pick for a fraction?

No — and proving this independence (well-definedness) is the technical heart of the construction. The elements of $\operatorname{Frac}(R)$ are equivalence classes $[\,(a,b)\,]$ , not pairs themselves, and the theorem's proof explicitly verifies that addition and multiplication give the same answer regardless of which representative pair is used, by a direct algebraic computation using the cross-multiplication relation. This is exactly analogous to checking that $1/2 + 1/3 = 5/6$ gives the same result whether you write $1/2$ as $1/2$ , $2/4$ , or $50/100$ .

Can the field of fractions be infinite even if we start with something small, like

R=\mathbb{Z}

Yes, and this is exactly the point: $\mathbb{Z}$ is a countably infinite integral domain, and $\operatorname{Frac}(\mathbb{Z}) = \mathbb{Q}$ is also countably infinite but strictly larger in the sense of containing multiplicative inverses $\mathbb{Z}$ lacks (e.g. $1/2 \notin \mathbb{Z}$ ). More strikingly, if $R$ is a finite integral domain, one can show $R$ must already be a field (a classical fact: injectivity of multiplication-by- $a$ for $a \neq 0$ , on a finite set, forces surjectivity, giving an inverse), so $\operatorname{Frac}(R) = R$ in the finite case, and genuinely new elements only appear when $R$ is infinite.

What goes wrong if we try to build

\operatorname{Frac}(R)

by inverting only some nonzero elements rather than all of them?

That is precisely the more general operation of localization: given a multiplicative subset $T\subseteq R\setminus\{0\}$ closed under products and containing $1$ , one forms $T^{-1}R$ using pairs $(a,t)$ with $t \in T$ , under the same cross-multiplication equivalence. This yields a ring (not always a field) sitting between $R$ and $\operatorname{Frac}(R)$ ; taking $T = R\setminus\{0\}$ recovers exactly the full field of fractions. Localization at a prime ideal's complement is a central tool in commutative algebra for studying rings "near a point," generalizing this construction far beyond fields.

Why is

\iota: R \to \operatorname{Frac}(R)

r \mapsto r/1

, automatically injective, and does this ever fail?

The proof shows $\iota(r)=\iota(s)$ forces $r\cdot 1_R = 1_R \cdot s$ , i.e. $r=s$ directly, with no cancellation of possible zero divisors needed — this step uses only that $1_R \ne 0_R$ and ordinary ring axioms, so it never fails once $R$ is an integral domain (in particular a nonzero ring). This is why $R$ genuinely "sits inside" $\operatorname{Frac}(R)$ as a subring, rather than merely mapping into it; without the integral domain hypothesis the whole target $\operatorname{Frac}(R)$ may not even be well-defined, as the counterexample demonstrates, so the question of injectivity of $\iota$ does not even arise.