The fundamental theorem of algebra

Theorem

The fundamental theorem of algebra

T-072Home MU-206Threads change · space

Statement

Let \( p(z) = a_n z^n + a_{n-1}z^{n-1} + \cdots + a_1 z + a_0 \) be a polynomial with coefficients \( a_0,\dots,a_n \in \mathbb{C} \), where \( n \geq 1 \) and \( a_n \neq 0 \) (so \( p \) is non-constant). Then there exists \( z_0 \in \mathbb{C} \) such that \( p(z_0) = 0 \). Equivalently, \( p \) factors completely over \( \mathbb{C} \) as \( p(z) = a_n \prod_{k=1}^{n} (z - z_k) \) for some (not necessarily distinct) \( z_1,\dots,z_n \in \mathbb{C} \).

Why it matters

This is the statement that closes off the process of extending number systems to solve polynomial equations: \( \mathbb{N} \to \mathbb{Z} \) to solve \( x + a = b \), \( \mathbb{Z} \to \mathbb{Q} \) to solve \( ax = b \), \( \mathbb{Q} \to \mathbb{R} \) to solve limits of Cauchy sequences (and \( x^2 = 2 \)), \( \mathbb{R} \to \mathbb{C} \) to solve \( x^2 = -1 \). The theorem says that this last step is the last step: \( \mathbb{C} \) is algebraically closed, and no further extension is needed to solve polynomial equations. Every real polynomial of odd degree already has a real root by the intermediate value theorem, but this theorem is the reason a real polynomial of any degree factors into linear and irreducible quadratic real factors — a fact used constantly in partial fractions, control theory, and ODE solving via characteristic polynomials.

Structurally, the theorem is also a striking advertisement for complex analysis: the cleanest proofs are not algebraic at all but come from a few lines of Liouville's theorem, illustrating how analytic machinery (holomorphicity, boundedness, growth estimates) settles a purely algebraic question that resisted centuries of purely algebraic attack.

Hypotheses

The coefficients lie in \( \mathbb{C} \) (or a field embeddable in \( \mathbb{C} \) with the same polynomial), and \( p \) is treated as a function on all of \( \mathbb{C} \). Over \( \mathbb{R} \) alone the statement is false: \( p(x) = x^2+1 \) has no real root, since \( x^2 + 1 \geq 1 \gt 0 \) for all \( x \in \mathbb{R} \). The theorem genuinely needs the codomain (and domain) to be \( \mathbb{C} \).

\( p \) is non-constant, i.e. \( \deg p = n \geq 1 \) (equivalently some coefficient with index \( \geq 1 \) is nonzero). The constant polynomial \( p(z) = 5 \) has no root at all — \( p(z) = 0 \) is never satisfied, so the degree bound cannot be dropped to \( n \geq 0 \).

\( p \) is a genuine polynomial (finite sum of powers of \( z \) with coefficients in \( \mathbb{C} \)), not merely a function that "looks algebraic," and it is evaluated on all of \( \mathbb{C} \), not a proper subset. The entire function \( f(z) = e^z \) is non-constant, holomorphic, and non-vanishing everywhere — showing that "non-constant + holomorphic on \( \mathbb{C} \)" alone (without the polynomial growth constraint of a finite-degree polynomial) is not enough for the conclusion; the proof genuinely uses that \( p(z) \to \infty \) as \( |z| \to \infty \) (false for \( e^z \) along the imaginary axis) or, algebraically, that \( p \) is a finite sum.

Proof

We give the classical analytic proof via Liouville's theorem. Fix a non-constant polynomial \( p(z) = a_n z^n + \cdots + a_0 \), \( n \geq 1 \), \( a_n \neq 0 \), and suppose for contradiction that \( p(z) \neq 0 \) for every \( z \in \mathbb{C} \).

\text{Define } f(z) = \dfrac{1}{p(z)}, \quad z \in \mathbb{C}.

Since \( p \) has no zeros in \( \mathbb{C} \) by assumption, \( f \) is well-defined everywhere on \( \mathbb{C} \), and \( f \) is holomorphic on \( \mathbb{C} \) as a quotient of holomorphic functions (polynomials are entire) with non-vanishing denominator — the quotient rule for holomorphic functions. A

|p(z)| \geq |a_n| |z|^n \left(1 - \sum_{k=0}^{n-1} \frac{|a_k|}{|a_n| |z|^{\,n-k}}\right)

Write \( p(z) = a_n z^n\left(1 + \dfrac{a_{n-1}}{a_n z} + \cdots + \dfrac{a_0}{a_n z^n}\right) \) for \( z \neq 0 \), and apply the reverse triangle inequality \( |A+B| \geq |A| - |B| \) repeatedly to the bracket. B

\exists\, R \gt 0 \text{ such that } |z| \geq R \implies \sum_{k=0}^{n-1} \frac{|a_k|}{|a_n||z|^{\,n-k}} \leq \frac{1}{2}

Each term \( \dfrac{|a_k|}{|a_n||z|^{n-k}} \to 0 \) as \( |z| \to \infty \) since \( n - k \geq 1 \); a finite sum of terms each tending to \( 0 \) tends to \( 0 \), so for the fixed target value \( \tfrac12 \) there is \( R \) beyond which the sum is below it — this is just the definition of the limit applied to a sum of \( n \) terms. A

|z| \geq R \implies |p(z)| \geq \tfrac12 |a_n| |z|^n \implies |f(z)| = \frac{1}{|p(z)|} \leq \frac{2}{|a_n|\,|z|^n} \xrightarrow[|z|\to\infty]{} 0

Substitute the bound from Step 3 into Step 2's inequality, then take reciprocals (both sides positive). This shows \( f(z) \to 0 \) as \( |z| \to \infty \); in particular \( f \) is bounded on \( \{ |z| \geq R \} \), say \( |f(z)| \leq M_1 \) there. B

\overline{D(0,R)} = \{\, z : |z| \leq R \,\} \text{ is compact, and } f \text{ is continuous, so } \exists\, M_2 : |f(z)| \leq M_2 \text{ on } \overline{D(0,R)}.

Holomorphic functions are continuous; a continuous real-valued function \( |f| \) on a compact set attains a maximum — the extreme value theorem, valid in \( \mathbb{C} \cong \mathbb{R}^2 \) since closed bounded sets are compact by Heine–Borel. A

|f(z)| \leq \max(M_1, M_2) =: M \quad \text{for all } z \in \mathbb{C}

Combine Steps 4 and 5: every \( z \in \mathbb{C} \) lies either in \( \overline{D(0,R)} \) or in \( \{|z| \geq R\} \), and \( f \) is bounded by \( M_1 \) or \( M_2 \) respectively on each region. So \( f \) is a bounded entire function. B

f \text{ is bounded and entire} \implies f \text{ is constant.}

Liouville's theorem (cited, standard result of MU-206): a bounded entire function is constant. Proof sketch for completeness: for any \( z_0 \in \mathbb{C} \) and \( r \gt 0 \), Cauchy's integral formula for the derivative gives \( f'(z_0) = \dfrac{1}{2\pi i}\displaystyle\oint_{|z-z_0|=r} \dfrac{f(z)}{(z-z_0)^2}\,dz \), so \( |f'(z_0)| \leq \dfrac{1}{2\pi} \cdot 2\pi r \cdot \dfrac{M}{r^2} = \dfrac{M}{r} \). Letting \( r \to \infty \) (legal since \( f \) is holomorphic, hence the formula holds for every \( r \), and \( f \) is bounded by \( M \) on the whole plane) forces \( f'(z_0) = 0 \). As \( z_0\) was arbitrary, \( f' \equiv 0 \) on the connected set \( \mathbb{C} \), so \( f \) is constant. C

f \equiv c \text{ constant} \implies p(z) = \frac{1}{c} \text{ for all } z \implies p \text{ is constant} \quad \text{(contradiction)}

If \( f = 1/p \) is constant and equal to \( c \), then necessarily \( c \neq 0 \) (else \( f \) would be identically \( \infty \), impossible for a genuine complex value), and \( p(z) = 1/c \) for every \( z\), i.e. \( p \) is the constant polynomial \( 1/c \). This contradicts the hypothesis \( \deg p = n \geq 1 \). A

\therefore \text{ the assumption "} p \text{ has no root" is false, i.e. } \exists\, z_0 \in \mathbb{C} : p(z_0) = 0.

Proof by contradiction (reductio ad absurdum): Steps 1–8 derived a false statement from the assumption that \( p \) is zero-free, so that assumption fails. A

p(z) = (z - z_0)\, q(z) \text{ for a polynomial } q \text{ of degree } n-1, \text{ and induction on } n \text{ gives the full factorisation.}

Polynomial division: since \( p(z_0) = 0 \), the factor theorem (division algorithm for polynomials over the field \( \mathbb{C} \): \( p(z) = (z-z_0)q(z) + r \) with \( r \) constant, and \( r = p(z_0) = 0 \)) gives \( p(z) = (z-z_0)q(z) \) with \( \deg q = n - 1 \) and leading coefficient still \( a_n \). If \( n - 1 \geq 1\), \( q\) is again non-constant, so Steps 1–9 apply to \( q \) to extract another root \( z_1\); repeating (formally, strong induction on \( n \), base case \( n=1\) where \( p(z) = a_1 z + a_0 = a_1(z - z_1)\) trivially) after exactly \( n \) extractions reduces \( q \) to a nonzero constant, namely \( a_n\), yielding \( p(z) = a_n(z-z_1)\cdots(z-z_n) \). C

Result

\forall\, p \in \mathbb{C}[z],\ \deg p \geq 1 \;\implies\; \exists\, z_0 \in \mathbb{C} : p(z_0) = 0

Reading. Any non-constant polynomial with complex coefficients has at least one complex root; consequently it splits completely into \( n = \deg p \) linear factors over \( \mathbb{C} \), counted with multiplicity.

Scope. Applies to every polynomial of degree \( \geq 1 \) with coefficients in \( \mathbb{C} \) (in particular, with real or rational or integer coefficients, since \( \mathbb{R}, \mathbb{Q}, \mathbb{Z} \subset \mathbb{C} \)). Does not apply, and is not claimed, for polynomials over fields not embedded in \( \mathbb{C} \) (e.g. finite fields \( \mathbb{F}_q \), or \( p \)-adic fields), which are algebraically closed only in their own separate algebraic closures.

Corollaries & converses

Complete factorisation. Every \( p \in \mathbb{C}[z] \) of degree \( n \) factors as \( a_n\prod_{k=1}^n (z-z_k) \); this is Step 10 of the proof, obtained by repeated application of the theorem plus the factor theorem.
\( \mathbb{C} \) is algebraically closed. By definition, a field is algebraically closed iff every non-constant polynomial over it has a root in it; the theorem is exactly the statement that \( \mathbb{C} \) satisfies this.
Real quadratic factorisation. Every \( p \in \mathbb{R}[z] \) factors over \( \mathbb{R} \) into a product of linear and irreducible quadratic factors, since non-real complex roots of a real polynomial occur in conjugate pairs \( z_0, \overline{z_0} \) (because \( p(\overline{z_0}) = \overline{p(z_0)} = 0 \) when coefficients are real), and \( (z-z_0)(z-\overline{z_0}) \) has real coefficients.
Converse. The converse — "if every non-constant polynomial with coefficients in a field \( F \) has a root in \( F \), then \( F \cong \mathbb{C} \)" — is false: \( \overline{\mathbb{Q}} \) (algebraic numbers) and \( \overline{\mathbb{F}_p} \) are also algebraically closed fields, neither isomorphic to \( \mathbb{C} \) (the first is countable, the second has positive characteristic). Algebraic closure is a property shared by many fields, not a characterisation of \( \mathbb{C} \) specifically.

Fails without

Dropping "coefficients/domain in \( \mathbb{C} \)": \( p(x) = x^2 + 1 \) over \( \mathbb{R} \) has no real root, since \( x^2 + 1 \geq 1 \) for all real \( x \); the theorem is specific to \( \mathbb{C} \).
Dropping "non-constant": \( p(z) = 7 \) (degree \( 0 \)) has no root anywhere, since \( p(z) - 0 = 7 \neq 0 \) always; boundedness of \( f = 1/p \) in the proof is trivial but there's no growth-to-zero step to force the contradiction, and indeed there's nothing to contradict — a nonzero constant genuinely never vanishes.
Dropping "finite-degree polynomial" (allowing power series / entire functions): \( f(z) = e^z \) is entire and non-constant but never zero (\( e^z = 0 \) has no solution since \( |e^z| = e^{\operatorname{Re}(z)} \gt 0 \) always). The proof's Step 2–4 growth estimate \( |p(z)| \to \infty \) fails for \( e^z\), which stays bounded along the imaginary axis, so \( 1/e^z \) is not shown bounded, and indeed is not.

Common errors

Believing the theorem gives a formula or algorithm for finding the root(s) — it is a pure existence statement; for \( n \geq 5\) no radical formula exists in general (Abel–Ruffini), yet FTA still guarantees roots exist (numerically approximable).
Confusing "algebraically closed" with "every polynomial has a real root" — students sometimes misapply FTA to real polynomials and expect a real root, forgetting the root guaranteed lies in \( \mathbb{C} \) and may be non-real.
Forgetting multiplicity: asserting a degree-\( n \) polynomial has \( n \) distinct roots. FTA only gives \( n \) roots counted with multiplicity; e.g. \( (z-1)^2 \) has the single root \( z=1\) with multiplicity 2.
Misapplying the Liouville-based proof to non-polynomial entire functions (e.g. claiming \( e^z \) "must" have a zero by the same argument) — the proof crucially uses that \( |p(z)| \to \infty \) as \( |z|\to\infty\), a property specific to (finite-degree, non-constant) polynomials, not entire functions generally.
Treating the "algebraic proof" (via odd-degree real polynomials plus induction on the 2-part of \( [\mathbb{C}:\mathbb{R}]\), using Galois theory and the fact that finite groups of odd order are solvable/have Sylow structure) as somehow more elementary than the analytic one — both are legitimate but each still imports a nontrivial external fact (IVT or Liouville).

Discussion

The theorem's name is something of a historical accident: it is not really a theorem "of algebra" in the sense of following from purely algebraic axioms — every known proof imports analysis (completeness of \( \mathbb{R} \), continuity, compactness, or Liouville's theorem) or topology (winding numbers, degree theory) at some point. Algebra alone, over an arbitrary field, cannot force existence of roots; what makes \( \mathbb{C} \) special is exactly its construction as the completion-and-algebraic-extension of \( \mathbb{R} \), a fundamentally analytic/topological fact smuggled into an algebraic-sounding statement. This is why every rigorous course proof, however "algebraic" in flavour, ultimately cites either the intermediate value theorem or a compactness/boundedness argument.

Historically, Gauss is credited with the first widely accepted proof (1799, his doctoral dissertation), though his proof had a topological gap (about the behaviour of algebraic curves) not rigorously closed until much later; he gave three further proofs over his lifetime, including an algebraic one using minimal analytic input. The Liouville-theorem proof given here is the standard 20th-century complex-analysis proof and is by far the shortest, reflecting how much of the "hard work" of the theorem gets absorbed into the (also nontrivial) machinery of Cauchy's integral formula.

There is also a purely topological proof via the argument principle / winding number: if \( p \) had no zero in a disc \( D(0,R) \) large enough that \( p \) is dominated by its leading term on \( |z|=R\), then the winding number of \( p(Re^{i\theta}) \) about \( 0 \) would have to simultaneously equal \( n \) (from the leading term \( a_n z^n\), which winds \( n\) times) and \( 0 \) (from continuously shrinking the contour to a point inside the zero-free disc, since a zero-free holomorphic image of a contractible loop has winding number \( 0 \)) — a contradiction unless \( n = 0\). This connects FTA to the broader machinery of degree theory in topology, and generalises (Brouwer degree) far beyond polynomials.

A subtler point often missed: the theorem is equivalent, in a precise sense, to the statement that the only finite field extension of \( \mathbb{R} \) besides \( \mathbb{R} \) itself is \( \mathbb{C} \) (Artin–Schreier theory characterises real-closed fields exactly via this "degree-2-closure" property). This connects FTA to a deep classification result: \( \mathbb{R} \) is essentially the unique complete ordered field, and \( \mathbb{C} = \mathbb{R}(i)\) is forced to be algebraically closed as a consequence of \( \mathbb{R}\) being real-closed — a fact that recasts FTA as a statement about \( \mathbb{R}\)'s order structure as much as about \( \mathbb{C}\)'s algebra. Common misconception: that FTA is "obvious" or purely computational because every polynomial equation one meets in a course happens to factor nicely — in fact the theorem asserts existence with no general constructive method for degree \( \geq 5\), a genuinely deep fact, not a computational triviality.

Worked examples

p(z) = z^4 + 1

A non-constant (\(n=4\)) polynomial with real (hence complex) coefficients; FTA guarantees 4 roots in \( \mathbb{C}\), counted with multiplicity. A

z^4 = -1 = e^{i\pi + 2\pi i k}, \quad k = 0,1,2,3

Write \(-1\) in polar form \( e^{i(\pi + 2\pi k)}\) for integer \(k\) (all representations of the same complex number), matching the algebraic equation \( z^4+1=0\) to \( z^4 = -1\). A

z_k = e^{i(\pi + 2\pi k)/4}, \quad k=0,1,2,3 \;\Longrightarrow\; z_0=e^{i\pi/4},\ z_1=e^{i3\pi/4},\ z_2=e^{i5\pi/4},\ z_3=e^{i7\pi/4}

De Moivre's formula for \( n\)-th roots: if \( z^n = re^{i\theta}\) then \( z = r^{1/n} e^{i(\theta+2\pi k)/n}\) for \( k=0,\dots,n-1\), giving exactly \( n\) distinct roots since \( r=1\) here. This exhibits explicitly the 4 roots FTA guaranteed abstractly. B

z^4+1 = \left(z-e^{i\pi/4}\right)\left(z-e^{i3\pi/4}\right)\left(z-e^{i5\pi/4}\right)\left(z-e^{i7\pi/4}\right)

Reading. All 4 roots are distinct (simple), confirming the degree-4 count from FTA, and pairing conjugates \( e^{i\pi/4},e^{-i\pi/4}=e^{i7\pi/4}\) and \( e^{i3\pi/4},e^{i5\pi/4}\) recovers the real factorisation \( z^4+1 = (z^2-\sqrt2 z+1)(z^2+\sqrt2 z+1)\).

p(z) = z^3 - 2z^2 + z - 2

Non-constant (\(n=3\)) real polynomial; FTA guarantees exactly 3 roots in \( \mathbb{C}\) (with multiplicity), and since \(n\) is odd there must be an odd number of real roots among them (as non-real roots pair up by conjugation, Corollary above). A

p(2) = 8 - 8 + 2 - 2 = 0

Trial evaluation at small integers (rational root theorem restricts candidates to divisors of the constant term \(-2\) over divisors of the leading coefficient \(1\): \( \pm1,\pm2\)); \(z=2\) works. A

p(z) = (z-2)(z^2+1) \quad \text{by polynomial division, since } p(2)=0

Factor theorem (used explicitly in Step 10 of the Proof section): \( p(z)/(z-2)\) has zero remainder because \( p(2)=0\); long division gives quotient \( z^2+1\). B

z^2+1=0 \implies z = \pm i

Solve the remaining quadratic factor directly; FTA (applied again to this degree-2 factor) guarantees these 2 roots exist in \( \mathbb{C}\), and here they are non-real, exhibiting a complex-conjugate pair as required by the real-coefficients corollary. A

z^3-2z^2+z-2 = (z-2)(z-i)(z+i)

Reading. Exactly one real root (\(z=2\)) and one non-real conjugate pair (\(\pm i\)), totalling 3 roots as FTA demands for a degree-3 polynomial, illustrating the odd-real-root pattern for odd-degree real polynomials.

Problems

State precisely why \( p(x) = x^4+1 \) having no real root does not contradict the Fundamental Theorem of Algebra.
Solution
FTA asserts a root exists in \( \mathbb{C}\), not in \( \mathbb{R}\); it makes no claim about real roots for real polynomials. Indeed \( x^4+1\gt 0\) for every real \(x\) (sum of a nonnegative term \(x^4\) and \(1\)), so it has no real root, but by Worked Example 1 it has 4 non-real complex roots \( e^{i\pi/4},e^{i3\pi/4},e^{i5\pi/4},e^{i7\pi/4}\), fully consistent with FTA.
Without finding the roots explicitly, explain why \( p(z) = z^5 - 3z + 1 \) (real coefficients) must have at least one real root.
Solution
By FTA, \(p\) has 5 roots in \( \mathbb{C}\) counted with multiplicity. Since coefficients are real, non-real roots occur in conjugate pairs \( \{z_0,\overline{z_0}\}\), each pair contributing 2 to the root count. Five is odd, and pairs contribute an even number, so the number of real roots (counted with multiplicity) must be odd, in particular at least 1. (One can alternatively see this directly via the Intermediate Value Theorem since \(p(-2)=-32+6+1=-25\lt0\) and \(p(2)=32-6+1=27\gt0\), but the question asks for the FTA-based argument.)
Give an explicit example of a field \(F\) and a non-constant polynomial in \(F[z]\) with no root in \(F\), to show the "coefficients/domain in \(\mathbb{C}\)" hypothesis is not vacuous, and identify exactly which step of the Liouville-theorem proof breaks down if one attempts to run it over \(F=\mathbb{R}\).
Solution
Take \(F=\mathbb{R}\), \(p(x)=x^2+1\); this has no root in \(\mathbb{R}\) since \(x^2+1\geq1\gt0\) for all real \(x\). Running the Liouville proof over \(\mathbb{R}\) fails already at Step 1: \(f=1/p\) would need to be shown "holomorphic," but holomorphicity (complex differentiability) is a notion for functions on (open subsets of) \(\mathbb{C}\); there is no real analogue giving the strong rigidity (Liouville's theorem, Cauchy's integral formula) that real-differentiable functions lack. The entire proof strategy is unavailable over \(\mathbb{R}\).
Prove, using FTA and the factor theorem, that a nonzero polynomial \(p\in\mathbb{C}[z]\) of degree \(n\) has at most \(n\) distinct roots.
Solution
By FTA (Result section) \(p(z)=a_n\prod_{k=1}^n(z-z_k)\) for some \(z_1,\dots,z_n\in\mathbb{C}\) (not necessarily distinct), \(a_n\neq0\). If \(w\) is any root of \(p\), then \(0=p(w)=a_n\prod_{k=1}^n(w-z_k)\); since \(\mathbb{C}\) is a field (no zero divisors) and \(a_n\neq0\), some factor \(w-z_k=0\), i.e. \(w=z_k\) for some \(k\in\{1,\dots,n\}\). So every root of \(p\) is one of the (at most \(n\)) values \(z_1,\dots,z_n\); hence \(p\) has at most \(n\) distinct roots.
Adapt the Liouville-theorem proof to show more generally: if \(g:\mathbb{C}\to\mathbb{C}\) is entire and there exist constants \(C\gt0\), \(R_0\gt0\), and integer \(n\geq1\) with \(|g(z)|\leq C|z|^n\) for all \(|z|\geq R_0\), then \(g\) is a polynomial of degree at most \(n\). (This is Liouville's theorem "for polynomials" — do not just cite it, sketch why it holds and note it is the converse growth direction to the FTA proof's Step 2–4.)
Solution
Since \(g\) is entire it has a global Taylor series \(g(z)=\sum_{m=0}^{\infty} c_m z^m\) valid on all of \(\mathbb{C}\), with \(c_m = \dfrac{1}{2\pi i}\displaystyle\oint_{|z|=r}\dfrac{g(z)}{z^{m+1}}\,dz\) by Cauchy's integral formula for Taylor coefficients, for every \(r\gt0\). For \(r\geq R_0\) and \(m\gt n\), the ML-inequality gives \(|c_m|\leq \dfrac{1}{2\pi}\cdot 2\pi r\cdot \dfrac{Cr^n}{r^{m+1}} = \dfrac{C}{r^{\,m-n}}\). Since \(m-n\gt0\), letting \(r\to\infty\) (legal, as the formula for \(c_m\) holds for every \(r\), by Cauchy's theorem applied on the entire, hence simply connected, domain \(\mathbb{C}\)) forces \(c_m=0\) for every \(m\gt n\). Hence \(g(z)=\sum_{m=0}^n c_m z^m\), a polynomial of degree at most \(n\). This is exactly the reverse implication to the growth estimate used in Steps 2–4 of the FTA proof: there, non-vanishing of a genuine degree-\(n\) polynomial produced an entire function bounded (degree 0 growth) and hence constant by the \(n=0\) case of this same fact; here the general statement shows any entire function of at most polynomial growth of degree \(n\) is a bona fide polynomial of that degree, the full strength of which the FTA proof only needed in its \(n=0\), "bounded \(\Rightarrow\) constant" special case (ordinary Liouville).