A permutation of a finite set looks like a harmless reordering until a formula starts to care about orientation. The determinant changes sign when two rows are swapped, an alternating tensor changes sign when two inputs are exchanged, and the roots of a polynomial split into orderings whose parity can carry field-theoretic information. This places signature directly between the group theory of symmetric and alternating groups, the linear algebra of determinants, and the Galois-theoretic action on polynomial roots. The signature of a permutation is the invariant that records whether a reordering preserves or reverses this orientation. The first problem is that a permutation has many descriptions. A cycle can be expanded into transpositions in more than one way, and extra cancelling swaps can be inserted without changing the permutation. If orientation is to be encoded by the number of swaps, then the parity of that number must be intrinsic rather than a feature of the chosen expression. [example: A Row Swap Changes Orientation] Let $V=\mathbb{R}^2$ with ordered basis $e_1=(1,0)$ and $e_2=(0,1)$. Here $S_2$ denotes the symmetric group on $\{1,2\}$, and let $\sigma\in S_2$ be the transposition exchanging $1$ and $2$. The ordered pair $(e_1,e_2)$ is represented by the column data first column $(1,0)$ and second column $(0,1)$, so the $2\times 2$ determinant formula gives \begin{align*} \det(e_1,e_2)=1\cdot 1-0\cdot 0=1. \end{align*} After applying the swap, the ordered pair becomes $(e_2,e_1)$, whose first column is $(0,1)$ and whose second column is $(1,0)$. Hence \begin{align*} \det(e_2,e_1)=0\cdot 0-1\cdot 1=-1. \end{align*} For this permutation, the only pair with $i<j$ is $(1,2)$, and $\sigma(1)=2>\sigma(2)=1$, so $\operatorname{Inv}(\sigma)=\{(1,2)\}$. Therefore \begin{align*} \operatorname{sgn}(\sigma)=(-1)^{|\operatorname{Inv}(\sigma)|}=(-1)^1=-1. \end{align*} Thus the swap changes the determinant by the factor $-1$, exactly matching the signature and reversing the orientation of the ordered basis. [/example] This example is small, but it contains the general phenomenon. Signature is not a measure of how far a permutation moves points. It measures whether the reordering reverses the alternating orientation seen by determinants, wedge products, and root-difference products. ## Definition Throughout this page, $n\in\mathbb{N}$ means $n\ge 1$, so all symmetric groups under discussion act on $\{1,\dots,n\}$. The page topic is the sign itself: a function from the symmetric group to the two possible orientation values. To define it without making arbitrary choices about transposition decompositions, we count order reversals directly. [definition: Signature] Let $n\in\mathbb{N}$. The signature, or sign, is the map \begin{align*} \operatorname{sgn}:S_n&\to\{1,-1\} \end{align*} defined by \begin{align*} \operatorname{sgn}(\sigma)=(-1)^{|\{(i,j):1\le i<j\le n,\ \sigma(i)>\sigma(j)\}|}. \end{align*} [/definition] A permutation with signature $1$ is called even, and a permutation with signature $-1$ is called odd. The definition uses the natural order on $\{1,\dots,n\}$, but its output will turn out to be an invariant of the group operation, not a fragile feature of the chosen notation. The exponent in the formula is doing the real work, so it needs its own language. Counting moved points is not enough: the list $3,5,1,4,2$ moves every symbol, while other permutations can move many symbols and still have either parity. What matters is whether a pair has crossed, meaning that two entries which originally appeared in increasing order now appear in decreasing order. Naming such a crossing as an inversion gives us a local unit of orientation reversal before we collect all reversals into the signature. [definition: Inversion of a Permutation] Let $n\in\mathbb{N}$ and let $\sigma\in S_n$. An inversion of $\sigma$ is a pair $(i,j)$ with $1\le i<j\le n$ such that $\sigma(i)>\sigma(j)$. [/definition] A single inversion is only a local crossing. To make later computations readable, we collect all such crossings into one finite set. [definition: Inversion Set] Let $n\in\mathbb{N}$. Here $\mathcal{P}(X)$ denotes the power set of a set $X$, meaning the set of all subsets of $X$. The inversion-set map is the function \begin{align*} \operatorname{Inv}:S_n&\to\mathcal{P}(\{(i,j):1\le i<j\le n\}) \end{align*} defined by \begin{align*} \operatorname{Inv}(\sigma)=\{(i,j):1\le i<j\le n,\ \sigma(i)>\sigma(j)\}. \end{align*} [/definition] With this notation, the defining formula becomes \begin{align*} \operatorname{sgn}(\sigma)=(-1)^{|\operatorname{Inv}(\sigma)|}. \end{align*} To see why this parity deserves to be called orientation, we need the elementary swaps from which all permutations are built. These swaps are the test case for every sign rule. [definition: Transposition] Let $n\in\mathbb{N}$. A transposition in $S_n$ is a permutation \begin{align*} \tau:\{1,\dots,n\}&\to\{1,\dots,n\} \end{align*} for which there exist distinct $a,b\in\{1,\dots,n\}$ such that $\tau(a)=b$, $\tau(b)=a$, and $\tau(k)=k$ for every $k\notin\{a,b\}$. [/definition] A transposition is the smallest possible non-identity reordering, but its effect on inversions is not limited to the two labels it exchanges. For instance, swapping two non-adjacent labels can reverse their order and also reverse the order of labels lying between them. The important point is therefore not the exact number of affected pairs, but the parity of that number: a single swap should reverse orientation exactly once modulo $2$. This creates the first local test for the inversion definition of sign. Before using transpositions as building blocks for arbitrary permutations, we must know that every individual transposition has the expected negative sign, even when the two exchanged positions are not adjacent. The obstruction is that a distant swap changes many pairwise comparisons at once, so its sign is not immediate from the definition. The next result isolates exactly this issue: it checks that all those reversed comparisons still contribute an odd number of inversions, making every transposition orientation-reversing. [quotetheorem:7875] This theorem is the first calibration point for the definition: it shows that the inversion count agrees with the geometric intuition from the row-swap example. It is also deliberately local. The result says what happens for one swap, but it does not yet say that every permutation can be built from such swaps or that the parity of a chosen build is independent of the build. If each elementary swap has negative sign, then any decomposition into swaps should determine the sign by the parity of its length. First we need the existence of such decompositions: transpositions are not only examples of odd permutations, but a generating family for the whole symmetric group. [quotetheorem:777] Generation gives a way to express every permutation using swaps, but it does not by itself make parity well-defined. The obstruction is non-uniqueness: the same permutation can have many transposition decompositions, and a sign computed from the number of swaps would be meaningless if those lengths could have different parity. [quotetheorem:7876] This result turns signature from a counting definition into an invariant of swap decompositions. The next question is whether it also respects multiplication of permutations. That property is what allows signature to define a homomorphism rather than only a label. [quotetheorem:7877] The homomorphism property explains why sign calculations can be broken into smaller pieces. For example, if $\sigma=(1\ 4\ 6\ 2)(3\ 8\ 5)$, then its sign can be computed as the product of the signs of the two disjoint cycles instead of by recounting every inversion in one-line notation. The theorem is also what makes the alternating group a kernel: without multiplicativity, the even permutations would be a parity class, but not automatically a subgroup. Its limitation is equally important: it simplifies products only after the factors are known, so the next section develops practical ways to recognize those factor signs from the notation in which a permutation is given. ## Computing Signature A definition is useful only if it can be computed in the forms in which permutations actually appear. Signature can be read from one-line notation, from transposition decompositions, from cycle notation, and from permutation matrices. Each method emphasizes a different aspect of the same invariant. ### Inversions in One-Line Notation The inversion definition is most direct when the permutation is written as the list $\sigma(1),\dots,\sigma(n)$. The calculation counts reversed pairs, so it trains the reader to see signature as order reversal. This method is especially useful when the permutation is already presented as a table or sequence. [example: Computing Signature by Inversions] Let $\sigma\in S_5$ be given in one-line notation by \begin{align*} \sigma(1)=3,\quad \sigma(2)=5,\quad \sigma(3)=1,\quad \sigma(4)=4,\quad \sigma(5)=2. \end{align*} We compute its inversion set by checking each pair $i<j$. For $i=1$, the values after $\sigma(1)=3$ are $5,1,4,2$, and the smaller ones are $1$ and $2$, so the inversions are $(1,3)$ and $(1,5)$. For $i=2$, the values after $\sigma(2)=5$ are $1,4,2$, and all three are smaller than $5$, so the inversions are $(2,3)$, $(2,4)$, and $(2,5)$. For $i=3$, the values after $\sigma(3)=1$ are $4$ and $2$, and neither is smaller than $1$, so there is no inversion starting at $3$. For $i=4$, we have $\sigma(4)=4>\sigma(5)=2$, so $(4,5)$ is an inversion. Thus \begin{align*} \operatorname{Inv}(\sigma)=\{(1,3),(1,5),(2,3),(2,4),(2,5),(4,5)\}. \end{align*} Hence \begin{align*} |\operatorname{Inv}(\sigma)|=6. \end{align*} By the definition of signature, \begin{align*} \operatorname{sgn}(\sigma)=(-1)^{|\operatorname{Inv}(\sigma)|}. \end{align*} Substituting the computed cardinality gives \begin{align*} \operatorname{sgn}(\sigma)=(-1)^6=1. \end{align*} Therefore $\sigma$ is even, even though every element of $\{1,2,3,4,5\}$ is moved by $\sigma$. [/example] This example rules out a common shortcut: the size of the support does not determine the sign. Pairwise order reversal is the relevant information. Transposition decompositions give another computational route. They are useful when a permutation has already been simplified into elementary swaps. The main danger is mistaking length for an invariant rather than parity of length. [example: Two Decompositions with the Same Parity] In $S_3$, compose permutations from right to left. We compute the product $(1\ 3)(1\ 2)$ by following each element through the rightmost transposition first. The element $1$ is sent by $(1\ 2)$ to $2$, and $(1\ 3)$ fixes $2$, so the product sends $1$ to $2$. The element $2$ is sent by $(1\ 2)$ to $1$, and then $(1\ 3)$ sends $1$ to $3$, so the product sends $2$ to $3$. The element $3$ is fixed by $(1\ 2)$, and then $(1\ 3)$ sends $3$ to $1$, so the product sends $3$ to $1$. Therefore \begin{align*} (1\ 3)(1\ 2)=(1\ 2\ 3). \end{align*} Now insert the cancelling pair $(1\ 2)(1\ 2)$. Since applying $(1\ 2)$ twice fixes $1$, fixes $2$, and fixes $3$, we have \begin{align*} (1\ 2)(1\ 2)=e. \end{align*} Thus \begin{align*} (1\ 2)(1\ 2)(1\ 3)(1\ 2)=e(1\ 3)(1\ 2). \end{align*} Since $e$ is the identity permutation, \begin{align*} e(1\ 3)(1\ 2)=(1\ 3)(1\ 2). \end{align*} Combining this with the first computation gives \begin{align*} (1\ 2)(1\ 2)(1\ 3)(1\ 2)=(1\ 2\ 3). \end{align*} The first decomposition uses $2$ transpositions, and the second uses $4$ transpositions. Both numbers are even, so both decompositions give the same parity contribution: \begin{align*} (-1)^2=1. \end{align*} Also, \begin{align*} (-1)^4=1. \end{align*} The inserted cancelling pair changes the length of the decomposition, but it changes it by $2$, so it does not change the resulting signature. [/example] The inserted pair changes the number of swaps but not its parity. This is the concrete content of the transposition-decomposition theorem. ### Cycles and Their Signs Cycle notation is the natural language of symmetric groups. A signature formula in terms of cycles avoids pairwise inversion counting and reveals how sign depends on cycle type. The first step is to understand a single cycle. [quotetheorem:7878] A cycle has positive signature when its length is odd and negative signature when its length is even. Conceptually, this reflects that an $m$-cycle can be assembled from $m-1$ transpositions, so only the parity of $m-1$ matters. Thus a $3$-cycle is even but a transposition is odd, and the result explains why the length of the orbit, rather than the particular symbols appearing in it, controls the sign. The limitation is that this theorem treats one cycle in isolation; a general permutation may move several disjoint sets of symbols at once, so we still need a rule that combines the separate cycle contributions without returning to inversion counting. Most permutations are best understood as products of disjoint cycles. The practical problem is to compute the sign without expanding every cycle into transpositions or recounting all inversions. Since [disjoint cycles commute](/theorems/774) and signature is multiplicative, the single-cycle formula should combine across the cycle decomposition. [quotetheorem:7879] This is usually the fastest hand computation of signature, but its importance is more structural than speed. The formula says that signature depends only on the cycle type: fixed points contribute length-$1$ factors and hence no sign change, while each nontrivial cycle contributes according to whether its length is even or odd. Equivalently, the total parity is controlled by how many even-length cycles occur in the [disjoint cycle decomposition](/theorems/775). This explains at once why all $3$-cycles lie in the alternating group and why a fixed point should not be counted as evidence for either parity; it has not participated in any reordering. The example below applies this cycle-type rule without expanding the permutation into transpositions or recounting inversions. [example: Signature from Cycle Type] Let $\sigma\in S_8$ be \begin{align*} \sigma=(1\ 4\ 6\ 2)(3\ 8\ 5). \end{align*} The disjoint cycle decomposition has one nontrivial cycle of length $4$ and one nontrivial cycle of length $3$; the element $7$ is fixed, so it contributes no nontrivial cycle factor. By *Signature from Disjoint Cycles*, \begin{align*} \operatorname{sgn}(\sigma)=(-1)^{4-1}(-1)^{3-1}. \end{align*} The exponents are \begin{align*} 4-1=3. \end{align*} Also, \begin{align*} 3-1=2. \end{align*} Therefore \begin{align*} \operatorname{sgn}(\sigma)=(-1)^3(-1)^2. \end{align*} Since $(-1)^3=-1$ and $(-1)^2=1$, this becomes \begin{align*} \operatorname{sgn}(\sigma)=(-1)\cdot 1=-1. \end{align*} Thus $\sigma$ is odd: the $4$-cycle contributes the negative factor, while the $3$-cycle contributes the positive factor. [/example] Cycle type is unchanged by relabelling the symbols. This gives a consistency check for signature: renaming the underlying labels should not alter whether a permutation is even or odd. Conjugation is exactly the group-theoretic operation that performs such a relabelling, so the sign of a permutation depends only on its cycle lengths, not on the particular symbols used to write those cycles. This is the relevant group-theoretic bridge to symmetric groups, alternating groups, normal subgroups, and later uses in Galois theory, where relabelling roots changes the displayed permutation without changing parity-dependent invariants such as the discriminant square class. This conjugacy invariance is a useful diagnostic. Any proposed sign formula that changes after a renaming of symbols cannot be the signature. ## The Alternating Group The kernel of a homomorphism is often the structure hidden inside an invariant. For signature, the kernel consists exactly of the orientation-preserving permutations. This subgroup is the alternating group, one of the central families in finite group theory and a natural continuation point after the symmetric group, cycle decompositions, kernels of homomorphisms, normal subgroups, and Galois actions on roots. If $f:G\to H$ is a [group homomorphism](/page/Group%20Homomorphism), then $\ker(f)$ denotes its kernel, the set of elements of $G$ mapped to the identity element of $H$. [definition: Alternating Group] Let $n\in\mathbb{N}$. The alternating group on $n$ letters is \begin{align*} A_n=\ker(\operatorname{sgn}:S_n\to\{1,-1\}). \end{align*} [/definition] Thus $A_n$ consists of the even permutations in $S_n$. The definition identifies $A_n$ as a kernel, so normality should follow from general group theory. We write $N\trianglelefteq G$ to mean that $N$ is a [normal subgroup](/page/Normal%20Subgroup) of $G$. The next issue is the size of $A_n$. Since any fixed transposition changes parity, even and odd permutations should occur in equal numbers when $n\ge 2$. [quotetheorem:7880] The size formula says that signature splits $S_n$ into two equal halves. The even half is not just a large subset; it is a normal subgroup. Small values of $n$ show how the construction begins before the large simple groups appear. These examples also prevent the reader from importing the $n\ge 5$ behavior into the low-dimensional cases. [example: The First Alternating Groups] For $n=1$, the only permutation is the identity $e$. Its inversion set is empty, so \begin{align*} \operatorname{sgn}(e)=(-1)^0=1. \end{align*} Thus $A_1=\ker(\operatorname{sgn})=S_1$. For $n=2$, the group $S_2$ consists of $e$ and $(1\ 2)$. The identity has no inversions, while $(1\ 2)$ sends $1$ to $2$ and $2$ to $1$, so the single pair $(1,2)$ is an inversion. Hence \begin{align*} \operatorname{sgn}(e)=1. \end{align*} Also, \begin{align*} \operatorname{sgn}((1\ 2))=(-1)^1=-1. \end{align*} Therefore \begin{align*} A_2=\{e\}. \end{align*} For $n=3$, list the six permutations by their inversion counts. The identity has no inversions, so it is even. The transposition $(1\ 2)$ has one-line notation $(2,1,3)$, so its only inversion is $(1,2)$. The transposition $(2\ 3)$ has one-line notation $(1,3,2)$, so its only inversion is $(2,3)$. The transposition $(1\ 3)$ has one-line notation $(3,2,1)$, so its inversions are $(1,2)$, $(1,3)$, and $(2,3)$. Thus all three transpositions are odd. For the cycle $(1\ 2\ 3)$, the one-line notation is $(2,3,1)$. The inversions are $(1,3)$ and $(2,3)$, so \begin{align*} \operatorname{sgn}((1\ 2\ 3))=(-1)^2=1. \end{align*} For the cycle $(1\ 3\ 2)$, the one-line notation is $(3,1,2)$. The inversions are $(1,2)$ and $(1,3)$, so \begin{align*} \operatorname{sgn}((1\ 3\ 2))=(-1)^2=1. \end{align*} Therefore \begin{align*} A_3=\{e,(1\ 2\ 3),(1\ 3\ 2)\}. \end{align*} If $g=(1\ 2\ 3)$, then $g^2=(1\ 3\ 2)$ and $g^3=e$, so these three elements are exactly $\langle g\rangle$. Hence $A_3$ is cyclic of order $3$. [/example] The alternating groups become much richer as $n$ grows. After passing from $S_n$ to the kernel of signature, the natural question is whether any further nontrivial normal structure remains inside the even permutations. For large enough degree, the answer is rigid enough to make $A_n$ a central object rather than an auxiliary construction. [quotetheorem:849] This result places signature at the entrance to the classification of finite simple groups. The map $S_n\to\{1,-1\}$ removes the only obvious normal subgroup of $S_n$, and the remaining even subgroup is simple for $n\ge 5$. ## Determinants and Alternating Forms Signature appears in linear algebra because determinants are alternating. Over coefficient fields where $1$ and $-1$ are distinct, swapping two columns reverses the determinant, and a general permutation multiplies it by the signature. This turns a group invariant into the sign convention behind determinant expansion; the characteristic-$2$ caveat is separated out later because there the abstract parity survives but its scalar realization collapses. ### Permutation Matrices To connect permutations with determinants, we represent a permutation by the matrix that permutes the standard basis. This representation gives a concrete [linear map](/page/Linear%20Map) whose determinant can be computed. The determinant then becomes a geometric way to read the same sign. [definition: Permutation Matrix] Let $n\in\mathbb{N}$ and let $\sigma\in S_n$. The permutation matrix associated to $\sigma$ is the matrix $P_\sigma\in\mathbb{R}^{n\times n}$ representing the linear map \begin{align*} P_\sigma:\mathbb{R}^n\to\mathbb{R}^n \end{align*} defined by $P_\sigma(e_j)=e_{\sigma(j)}$ for each standard basis vector $e_j\in\mathbb{R}^n$. [/definition] The matrix $P_\sigma$ implements the relabelling of basis vectors by $\sigma$. The question is whether the abstract parity invariant agrees with the linear-algebraic test for orientation, namely the determinant. This comparison is what lets sign computations move between permutations and matrices. [quotetheorem:7881] This theorem gives a second way to recognize parity: build the linear map that relabels the standard basis and take its determinant. The point is not only computational. It says that the abstract [sign homomorphism](/theorems/778) agrees with the orientation test already present in linear algebra, so a permutation reverses orientation exactly when its permutation matrix has determinant $-1$. The limitation is that this statement uses the standard permutation matrix representation; it does not by itself explain why arbitrary determinant expansions require signs for all possible row-column matchings. That broader mechanism is supplied by the Leibniz formula. Determinants must sum over all possible ways of selecting one entry from each row and each column. Without a sign assigned to each matching, the alternating cancellation property would be lost. The next formula shows how signature supplies precisely those coefficients. [quotetheorem:7882] The formula is often the first place signature appears in linear algebra. It also explains why determinant terms cancel when rows or columns repeat. [example: Cancellation When Two Rows Agree] Let $A\in\mathbb{R}^{2\times 2}$ have both rows equal to $(a,b)$, so \begin{align*} A_{11}=a,\quad A_{12}=b,\quad A_{21}=a,\quad A_{22}=b. \end{align*} The group $S_2$ has exactly two elements: the identity $e$, with $e(1)=1$ and $e(2)=2$, and the transposition $\tau=(1\ 2)$, with $\tau(1)=2$ and $\tau(2)=1$. By *[Leibniz Formula for the Determinant](/theorems/7882)*, \begin{align*} \det A=\operatorname{sgn}(e)A_{1,e(1)}A_{2,e(2)}+\operatorname{sgn}(\tau)A_{1,\tau(1)}A_{2,\tau(2)}. \end{align*} Since $e$ has no inversions, $\operatorname{sgn}(e)=(-1)^0=1$. Since $\tau$ has the single inversion $(1,2)$, $\operatorname{sgn}(\tau)=(-1)^1=-1$. Substituting the entries of $A$ gives \begin{align*} \det A=1\cdot A_{11}A_{22}+(-1)\cdot A_{12}A_{21}. \end{align*} Hence \begin{align*} \det A=1\cdot a b+(-1)\cdot b a. \end{align*} Because multiplication in $\mathbb{R}$ is commutative, $ba=ab$, so \begin{align*} \det A=ab-ab=0. \end{align*} The two determinant terms choose the same product from the repeated rows, and their opposite signatures make the two contributions cancel. [/example] This cancellation is not special to dimension two. It is the basic behavior of alternating multilinear forms, where signature controls the effect of permuting inputs. ### Alternating Multilinear Forms Determinants are only one example of a broader construction. In many algebraic settings, we want a multilinear function that vanishes when two inputs coincide and changes sign when two inputs are swapped. The following definition isolates the vanishing condition, which remains meaningful over any field. [definition: Alternating Multilinear Form] Let $F$ be a field, let $V$ be a [vector space](/page/Vector%20Space) over $F$, and let $m\in\mathbb{N}$. An alternating $m$-linear form on $V$ is a map $\omega:V^m\to F$ that is linear in each argument and satisfies \begin{align*} \omega(v_1,\dots,v_m)=0 \end{align*} whenever $v_i=v_j$ for some distinct $i,j\in\{1,\dots,m\}$. [/definition] The definition uses repeated inputs because that condition survives in all characteristics. In fields where $1$ and $-1$ are distinct, this vanishing condition should force a sign change whenever arguments are permuted. The point now is to identify the precise scalar attached to a permutation of the inputs: it should depend only on the permutation, not on the particular vectors or on the form. The quoted theorem formulates that scalar as the sign of the permutation and uses its own integer degree variable independently of the field $F$ above. [quotetheorem:3556] This theorem is the algebraic heart of orientation. It explains why the alternating condition is strong enough, in the intended setting where signs are meaningful, to make every reordering act by a predictable scalar. The hypothesis is not cosmetic: in characteristic two, the distinction between even and odd signs collapses because $1=-1$, so the alternating condition cannot encode the same orientation data. For example, swapping two columns of a determinant should reverse its value; the theorem says that the same sign rule is forced for every alternating multilinear form, not just for the determinant. This is the sign convention behind exterior powers, wedge products, and the determinant as a volume form, and it prepares the next use of signature as a way to measure how permutations act on ordered algebraic data. ## Polynomial Discriminants and Galois Actions Signature also appears when permutations act on roots of a polynomial. To detect whether a root permutation is even or odd, we need an expression that is almost symmetric in the roots but still changes sign when two roots are exchanged. The product of all pairwise root differences has exactly this ordering dependence: reversing two roots reverses the orientation encoded by the product, while squaring it removes the sign and produces the discriminant. The notation in this section is standard field notation. A [field extension](/page/Field%20Extension) $E/k$ means that $E$ is a field containing a copy of $k$. An [algebraic closure](/page/Algebraic%20Closure) $\overline{k}$ is a field containing all algebraic roots over $k$, and $k[x]$ denotes the [polynomial ring](/page/Polynomial%20Ring) in the variable $x$ with coefficients in $k$. A product of all pairwise root differences remembers an ordering of the roots. The formal object below packages that orientation-sensitive product so that the action of a permutation on the roots can be measured by a single scalar factor. [definition: Vandermonde Product] Let $k$ be a field, let $E/k$ be a field extension, and let $n\in\mathbb{N}$. The Vandermonde product is the map \begin{align*} \Delta:E^n&\to E \end{align*} defined by \begin{align*} \Delta(\alpha_1,\dots,\alpha_n)=\prod_{1\le i<j\le n}(\alpha_j-\alpha_i). \end{align*} [/definition] The Vandermonde product is not symmetric, so it is sensitive to how the roots are ordered. The obstruction is that a permutation of the roots rearranges many factors at once, making the sign change less visible than in a single displayed difference. The point is to show that all those factor rearrangements combine to exactly the signature of the permutation. [quotetheorem:7883] Squaring the Vandermonde product removes the sign. That creates an invariant of the unordered set of roots, which is the kind of quantity that can be recovered from the coefficients of the polynomial. The remaining task is to package this symmetric root expression as a function of the polynomial itself, not of a chosen ordering of its roots in an algebraic closure. The following definition does this by taking the square of the Vandermonde product and adding the correct power of the leading coefficient, so the invariant applies uniformly to non-monic degree-$n$ polynomials. [definition: Discriminant of a Polynomial] Let $k$ be a field, let $n\in\mathbb{N}$, and fix an algebraic closure $\overline{k}$ of $k$. Let $\mathcal{P}_n(k)$ be the set of degree-$n$ polynomials in $k[x]$. The discriminant is the map \begin{align*} \operatorname{disc}:\mathcal{P}_n(k)&\to k \end{align*} defined as follows: if \begin{align*} f(x)=a\prod_{i=1}^n(x-\alpha_i) \end{align*} with $a\in k\setminus\{0\}$ and $\alpha_1,\dots,\alpha_n\in\overline{k}$, then \begin{align*} \operatorname{disc}(f)=a^{2n-2}\prod_{1\le i<j\le n}(\alpha_i-\alpha_j)^2. \end{align*} [/definition] The discriminant is symmetric, but the unsquared Vandermonde product still remembers parity. Separability is the hypothesis that makes this memory nondegenerate: the roots are distinct, so the Vandermonde product is nonzero and its square is the discriminant up to the leading-coefficient factor. We use the sign convention for the Vandermonde product that matches the theorem below, namely $\prod_{1\le i<j\le n}(\alpha_j-\alpha_i)$; replacing it by $\prod_{1\le i<j\le n}(\alpha_i-\alpha_j)$ changes only the global factor $(-1)^{n(n-1)/2}$, so its square and the square criterion are unchanged. When an element of the [Galois group](/page/Galois%20Group) permutes the roots, it multiplies the Vandermonde product by the signature of that permutation. The characteristic restriction is essential here: only in characteristic not $2$ can the signs $1$ and $-1$ distinguish even and odd permutations. Under that hypothesis, fixing the Vandermonde product is equivalent to acting by an even permutation, and the following theorem packages this observation as a square test for the discriminant. [remark: Discriminant square test] For the square criterion used here, assume throughout this paragraph that the base field has characteristic not $2$. Under this hypothesis, the discriminant of a separable polynomial is a square in the base field exactly when every Galois automorphism acts on the roots by an even permutation, so the Galois group is contained in the alternating group. This is the characteristic-not-$2$ form of the discriminant-and-alternating-group criterion. [/remark] This criterion turns an order-parity question into a field-theoretic square test, with the working hypothesis made explicit that the base field has characteristic not $2$. The separability assumption keeps the Vandermonde product nonzero, while the characteristic assumption makes its sign meaningful; without either hypothesis, the alternating-group conclusion no longer reflects the same parity information. It is one of the standard reasons signature is useful beyond elementary permutation calculations. [example: A Cubic with Non-Square Discriminant] Let $f(x)=x^3-2\in\mathbb{Q}[x]$. Since $f$ has the monic cubic form $x^3+px+q$ with $p=0$ and $q=-2$, the cubic discriminant formula gives \begin{align*} \operatorname{disc}(f)=-4p^3-27q^2. \end{align*} Substituting $p=0$ and $q=-2$ gives \begin{align*} \operatorname{disc}(f)=-4\cdot 0^3-27\cdot(-2)^2. \end{align*} Now $0^3=0$ and $(-2)^2=4$, so \begin{align*} \operatorname{disc}(f)=-4\cdot 0-27\cdot 4. \end{align*} Thus \begin{align*} \operatorname{disc}(f)=0-108=-108. \end{align*} The rational number $-108$ is not a square in $\mathbb{Q}$, because every rational square $r^2$ is nonnegative in the usual ordering of $\mathbb{Q}$. Also, $f$ is separable over $\mathbb{Q}$: its derivative is $f'(x)=3x^2$, and $x=0$ is the only root of $f'$, while $f(0)=-2\ne 0$. Since $\mathbb{Q}$ has characteristic not $2$, the discriminant square test applies. Therefore the Galois group of $x^3-2$ acting on its three roots is not contained in $A_3$. [/example] The example shows how a numerical invariant of a polynomial can detect odd root permutations. In degree three, that distinction already narrows the possible Galois group substantially. ## Characteristic Two and Failure Modes Signature depends on the distinction between the two signs. As an abstract homomorphism $S_n\to\{1,-1\}$ it exists for every $n$, but determinant-valued realizations can lose the scalar distinction when the coefficient field has characteristic $2$. The determinant of a permutation matrix usually realizes the signature as a scalar. Over a field where $1_k=-1_k$, this scalar realization cannot distinguish even and odd permutations. The abstract group invariant remains present, but the field no longer sees the two values separately. [example: Sign Disappears in Characteristic $2$ Determinants] Let $k$ be a field with $1_k+1_k=0_k$, and let $P\in k^{2\times 2}$ be the matrix that exchanges the two standard basis vectors $e_1$ and $e_2$. Since $P(e_1)=e_2$ and $P(e_2)=e_1$, the entries of $P$ are \begin{align*} P_{11}=0_k,\quad P_{12}=1_k,\quad P_{21}=1_k,\quad P_{22}=0_k. \end{align*} Using the $2\times 2$ determinant formula, \begin{align*} \det P=P_{11}P_{22}-P_{12}P_{21}. \end{align*} Substituting the four entries gives \begin{align*} \det P=0_k\cdot 0_k-1_k\cdot 1_k. \end{align*} Since $0_k\cdot 0_k=0_k$ and $1_k\cdot 1_k=1_k$, we get \begin{align*} \det P=0_k-1_k=-1_k. \end{align*} The characteristic-$2$ hypothesis gives $1_k+1_k=0_k$. By definition of additive inverse, $1_k+(-1_k)=0_k$. Since additive inverses are unique, $-1_k=1_k$. Therefore \begin{align*} \det P=-1_k=1_k. \end{align*} The underlying permutation is the transposition exchanging $1$ and $2$, whose inversion set is $\{(1,2)\}$, so its abstract signature is $(-1)^1=-1$. Thus the transposition is still odd as a permutation, but its determinant over $k$ is the scalar $1_k$, so the characteristic-$2$ determinant no longer distinguishes the two signs. [/example] This is not a contradiction. It separates the group-theoretic signature from its scalar image in a particular coefficient field. The same issue appears for alternating multilinear forms. If sign changes are invisible, then defining alternating behavior only by skew-symmetry can lose information. The vanishing-on-repeated-inputs definition avoids this problem. [remark: Characteristic $2$] In characteristic $2$, the equality $-1_k=1_k$ means that skew-symmetry and symmetry can coincide as scalar identities. For this reason, alternating forms are usually defined by the condition that the form vanishes whenever two arguments are equal, rather than only by a sign-change rule under transpositions. [/remark] Characteristic $2$ is a useful warning. Signature is universal as a group invariant, but not every algebraic environment can realize its two values as distinct scalars. ## Beyond and Connected Topics Signature is the entry point to several larger constructions in algebra. The most immediate continuation is the study of symmetric and alternating groups. The alternating group $A_n$ is the kernel of the [signature homomorphism](/theorems/7877), and for $n\ge 5$ it becomes a basic source of finite simple groups. A second continuation is multilinear algebra. Determinants, alternating forms, and exterior powers all encode the same sign rule. This is why a permutation of wedge factors contributes $\operatorname{sgn}(\sigma)$ and why orientation in finite-dimensional vector spaces is fundamentally a statement about even and odd changes of ordered bases. A third continuation is Galois theory. Once a polynomial's Galois group is viewed as a subgroup of $S_n$, signature gives a quadratic character of that group. The discriminant tells whether this character is trivial, turning root-order parity into a field-theoretic condition. A fourth connection appears in coding and cryptography. Permutations are used to rearrange coordinates, symbols, and states; signature is rarely the whole invariant, but it is a fast test for whether a permutation lies in the even subgroup. This matters when a construction restricts to $A_n$ or when parity constraints are part of a permutation-based protocol. The linear algebra background for determinants and alternating forms is developed in [Cambridge IB Linear Algebra](/page/Cambridge%20IB%20Linear%20Algebra). Applications involving finite fields and permutation actions connect naturally to [Cambridge II Coding and Cryptography](/page/Cambridge%20II%20Coding%20and%20Cryptography). ## References Androma, [Cambridge IB Linear Algebra](/page/Cambridge%20IB%20Linear%20Algebra). Androma, [Cambridge II Coding and Cryptography](/page/Cambridge%20II%20Coding%20and%20Cryptography). Dummit and Foote, *Abstract Algebra* (2004). Artin, *Algebra* (2011). Lang, *Algebra* (2002).