Trace is the operation that extracts the additive diagonal content of a linear transformation. For a matrix it is the sum of diagonal entries, but the real point is basis-independence: trace is attached to the [linear map](/page/Linear%20Map), not to a particular coordinate display. This makes it a bridge between [matrix](/page/Matrix) computations, [eigenvalue and eigenvector](/page/Eigenvalue%20and%20Eigenvector) information, [module](/page/Module) theory, and [field extension](/page/Field%20Extension) arithmetic. In linear algebra, trace records the first-order effect of an endomorphism on volume through determinants such as $\det(I+tA)$: for an $n \times n$ matrix $A$ over a field, \begin{align*} \det(I+tA)=1+t\operatorname{tr}(A)+\text{terms of degree at least }2\text{ in }t \end{align*} as a polynomial identity. In representation theory, characters are traces of representing matrices. In algebraic number theory, the field trace turns multiplication by an element into a scalar in the base field, producing trace forms, discriminants, and dual bases. The same word is used because all of these constructions reduce an endomorphism to a scalar in a way compatible with change of basis and cyclic composition. ## Definition ### Linear and Matrix Trace A linear transformation may be written by many different matrices, so the central object is not the array of entries but the scalar that survives every change of basis. The trace is first of all this coordinate-independent scalar attached to an endomorphism; the familiar diagonal sum is the computational formula used after a basis has been chosen. [definition: Trace of a Linear Endomorphism] Let $k$ be a field and let $V$ be a finite-dimensional $k$-[vector space](/page/Vector%20Space) with $\dim_k V=n$. The trace of a linear endomorphism is the map \begin{align*} \operatorname{tr}: \operatorname{End}_k(V) \to k. \end{align*} It sends $T$ to the scalar \begin{align*} \sum_{i=1}^n A_{ii}, \end{align*} where $A \in k^{n \times n}$ is the matrix of $T$ with respect to any basis of $V$. [/definition] To prove and reuse this invariant, we need a separate name for the raw diagonal-sum operation on a square matrix. This matrix-level definition is the computational object whose behaviour under similarity makes the operator-level definition well-defined. [definition: Matrix Trace] Let $R$ be a commutative ring and let $n \in \mathbb{N}$. The matrix trace is the map \begin{align*} \operatorname{tr}: R^{n \times n} \to R. \end{align*} It sends $A$ to \begin{align*} \sum_{i=1}^n A_{ii}. \end{align*} [/definition] A first computation shows how little of a displayed matrix trace actually uses. Only diagonal entries enter the scalar, even though off-diagonal entries can still affect eigenvectors and the geometry of the map. [example: Trace of a Matrix] Let $A \in \mathbb{R}^{3 \times 3}$ have diagonal entries $A_{11}=2$, $A_{22}=3$, and $A_{33}=-6$. By the definition of matrix trace, \begin{align*} \operatorname{tr}(A)=A_{11}+A_{22}+A_{33}=2+3+(-6)=-1. \end{align*} Changing only an off-diagonal entry, such as replacing $A_{12}$ by $100$, does not change this trace value. The trace is therefore not the sum of all matrix entries; it is the diagonal sum. [/example] A coordinate formula only becomes an invariant if it survives change of coordinates. A linear map can be represented by many similar matrices, so the diagonal sum would be useless as an operator invariant if similar matrices could have different traces. The next result supplies the missing invariance test. It isolates the basis-change operation as matrix conjugation and shows that trace is unchanged under that operation, so trace can be used as a scalar attached to the underlying linear map rather than to a chosen matrix. [quotetheorem:7811] This invariance theorem is the reason the diagonal formula can be promoted from a matrix recipe to a definition for linear maps. Once similar matrices have the same trace, choosing a different basis changes the representing matrix but not the scalar being measured. It also explains why characters in representation theory are constant on conjugacy classes: conjugating the representing matrix leaves its trace unchanged. The result has a precise limitation. It says trace is unchanged under similarity, not that trace determines the similarity class. Many non-similar matrices can have the same trace, so trace is a coarse invariant rather than a complete classifier. Its usefulness comes from being easy to compute, stable under basis change, and compatible with the algebraic operations developed below. ### Module and Field Trace Algebra often asks the same linear questions over rings rather than fields. For a finite free module, an endomorphism still has a matrix after a basis is chosen, but the resulting scalar must not depend on which basis was used. Similarity-invariance of matrix trace makes it possible to define trace directly for such module endomorphisms. [definition: Trace of an Endomorphism of a Finite Free Module] Let $R$ be a commutative ring and let $M$ be a finite free $R$-module. The trace of an endomorphism of $M$ is the map \begin{align*} \operatorname{tr}: \operatorname{End}_R(M) \to R. \end{align*} It sends $T$ to \begin{align*} \operatorname{tr}(A), \end{align*} where $A$ is the matrix of $T$ with respect to any $R$-basis of $M$. [/definition] A field extension creates a vector space over its base field, so every element can act by multiplication. The next definition is needed because it turns an element of the larger field into a scalar in the smaller field. This is the trace used in Galois theory and algebraic number theory. [definition: Field Trace] Let $K/k$ be a finite field extension. For $\alpha \in K$, let $m_\alpha: K \to K$ be the $k$-linear map defined by $m_\alpha(x)=\alpha x$. The field trace from $K$ to $k$ is the map \begin{align*} \operatorname{Tr}_{K/k}: K \to k. \end{align*} It sends $\alpha$ to \begin{align*} \operatorname{tr}(m_\alpha). \end{align*} [/definition] Quadratic extensions are the first place where the field trace becomes a practical computation rather than only a definition. They show that the trace can discard the irrational part of an element while retaining the part visible from the base field. [example: Trace in a Quadratic Extension] Let $K=\mathbb{Q}(\sqrt{2})$ and $k=\mathbb{Q}$. For $\alpha=a+b\sqrt{2}$ with $a,b\in\mathbb{Q}$, compute the $\mathbb{Q}$-linear map $m_\alpha:K\to K$ in the basis $(1,\sqrt{2})$. Since $(\sqrt{2})^2=2$, \begin{align*} m_\alpha(1)=(a+b\sqrt{2})\cdot 1=a+b\sqrt{2}. \end{align*} Also, \begin{align*} m_\alpha(\sqrt{2})=(a+b\sqrt{2})\sqrt{2}=a\sqrt{2}+b(\sqrt{2})^2=a\sqrt{2}+2b=2b+a\sqrt{2}. \end{align*} Thus the first basis vector is sent to $a\cdot 1+b\cdot\sqrt{2}$, and the second basis vector is sent to $2b\cdot 1+a\cdot\sqrt{2}$. Therefore the diagonal coefficients of the representing matrix are $a$ and $a$, so by the definition of field trace, \begin{align*} \operatorname{Tr}_{K/k}(a+b\sqrt{2})=\operatorname{tr}(m_\alpha)=a+a=2a. \end{align*} The conjugate calculation gives the same scalar: \begin{align*} (a+b\sqrt{2})+(a-b\sqrt{2})=a+a+b\sqrt{2}-b\sqrt{2}=2a. \end{align*} In this quadratic extension, the trace keeps twice the rational part and cancels the $\sqrt{2}$-part. [/example] This example explains why the trace is useful in arithmetic: it turns an element of $K$ into a scalar of $k$ without choosing an ordering, embedding, or preferred conjugate. Field trace becomes more powerful when it is used to compare two elements at once. A single trace value records one element, but many arithmetic questions require testing how two elements interact under multiplication. Tracing the product turns multiplication in the extension into a scalar-valued bilinear measurement on the base field. [definition: Trace Pairing] Let $K/k$ be a finite field extension. The trace pairing is the map \begin{align*} B_{K/k}: K \times K \to k. \end{align*} It sends $(\alpha,\beta)$ to \begin{align*} \operatorname{Tr}_{K/k}(\alpha\beta). \end{align*} [/definition] To treat the trace pairing as a [bilinear form](/page/Bilinear%20Form) rather than only a formula, we need to know which formal symmetries it has. Symmetry is the first test: it says the pairing does not remember which factor was placed on the left before multiplication. [quotetheorem:7812] Symmetry is what lets the trace pairing behave like a bilinear form rather than as a merely ordered multiplication test. Once $B_{K/k}(\alpha,\beta)=B_{K/k}(\beta,\alpha)$ is known, orthogonality, radicals, and discriminant matrices can be discussed without keeping track of a left input and a right input separately. The proof uses only commutativity of multiplication in the field, so it is a formal property of this particular pairing, not a separability theorem. The limitation is just as important: symmetry by itself says nothing about whether the pairing detects nonzero elements. A symmetric bilinear form can still be identically zero, as purely inseparable examples show. The question of whether this pairing is degenerate therefore becomes a separate central separability test later in the page. ## Equivalent Characterisations ### Characteristic-Polynomial Formula The diagonal-sum formula is often the fastest way to compute trace, but it does not reveal why trace is tied to eigenvalues and characteristic polynomials. The characteristic polynomial records the determinant of $tI-A$, and its next-to-leading coefficient is exactly where trace appears. [definition: Characteristic Polynomial] Let $k$ be a field and let $V$ be a finite-dimensional $k$-vector space with $\dim_k V = n$. The characteristic polynomial construction is the map \begin{align*} \chi: \operatorname{End}_k(V) \to k[t]. \end{align*} Here $k[t]$ denotes the [polynomial ring](/page/Polynomial%20Ring) in one variable $t$ over $k$. The construction sends $T$ to \begin{align*} \det(tI - A), \end{align*} where $A$ is the matrix of $T$ with respect to any basis of $V$. [/definition] Trace should be recoverable from invariant data, not only from a displayed matrix. The characteristic polynomial is basis-independent, so its coefficients provide a coordinate-free place where trace can appear. The coefficient comparison below identifies exactly where trace sits inside the characteristic polynomial. This is the bridge from the diagonal-sum definition to later statements that compute trace from eigenvalues and field-theoretic conjugates. [quotetheorem:7813] The coefficient formula gives a coordinate-free way to find trace even when no particular basis is convenient. Instead of summing a displayed diagonal, one can read trace from the characteristic polynomial, which is unchanged by similarity. This is why trace travels together with determinant and eigenvalue multiplicities as part of the same invariant package. The statement should not be read backward as saying that trace determines the characteristic polynomial. It identifies one coefficient, while the remaining coefficients contain additional information such as determinant and higher elementary symmetric functions of eigenvalues. The next theorem adds the eigenvalue interpretation of this same coefficient. ### Eigenvalues and Field Elements The next theorem is needed because eigenvalues may exist only after extending the field, yet their symmetric sums still produce scalars in the original field. It explains the phrase that trace is the sum of eigenvalues. It also shows why trace is insensitive to diagonalizability. [quotetheorem:922] This theorem is the familiar slogan "trace is the sum of eigenvalues" in a form that survives [algebraic closure](/page/Algebraic%20Closure) and repeated roots. It is useful because eigenvalues often reveal the action of an operator more transparently than matrix entries, while the symmetric sum still lands in the original base field. In computations, this lets one switch between diagonal pictures, characteristic polynomials, and coordinate-free trace. The limitation is that trace records only the first symmetric sum of the eigenvalues. It does not determine the full spectrum, the determinant, or the Jordan structure of an operator. For field trace, however, the eigenvalue perspective points toward conjugates of elements in finite extensions. For field extensions, the multiplication map $m_\alpha$ has eigenvalues related to the possible images of $\alpha$ in an algebraic closure. These images depend on embeddings over the base field, not on arbitrary choices of coordinates in $K$. To state trace formulas in those terms, we need a precise name for the field-theoretic counterparts of eigenvalues. [definition: Conjugates in a Finite Extension] Let $K/k$ be a finite extension and let $\alpha \in K$. The conjugates of $\alpha$ over $k$ are the images $\sigma(\alpha)$ as $\sigma$ ranges over the $k$-embeddings from $k(\alpha)$ into an algebraic closure $\overline{k}$. [/definition] The next theorem is needed because an element $\alpha \in L$ may generate only the intermediate field $K(\alpha)$, not all of $L$. Its [minimal polynomial](/page/Minimal%20Polynomial) over $K$ still records the trace and the field norm of multiplication by $\alpha$, where $\operatorname{N}_{L/K}(\alpha)$ denotes the determinant of the $K$-linear multiplication map $m_\alpha:L\to L$. The coefficients must be counted with the extension multiplicity $r=[L:K(\alpha)]$. This gives a coefficient formula that works directly from $P_\alpha$, without choosing a basis of $L$ or listing embeddings. [quotetheorem:1292] The formula says that field trace and norm are controlled by the first and last coefficients of the minimal polynomial, corrected by the factor $r$ measuring how much larger $L$ is than the field generated by $\alpha$. When $\alpha$ generates the whole extension, $r=1$ and the familiar coefficient rules are recovered. When it does not, each eigenvalue of multiplication by $\alpha$ occurs with multiplicity $r$, which is why both the trace coefficient and the norm coefficient are adjusted. The limitation is that this theorem computes the trace and norm of a single element through its minimal polynomial; it is not an embedding-count formula for arbitrary separable extensions. Separability gives a different interpretation in terms of conjugates, while the multiplication-map definition remains valid in every finite extension. ## Computations and Pathologies ### Diagonal Operators A diagonal matrix makes the relationship with eigenvalues visible without any polynomial calculation. It is the model case behind the general eigenvalue formula. [example: Trace of a Diagonalizable Operator] Let $(v_1,v_2)$ be a basis of $\mathbb{R}^2$ in which $T$ is represented by the diagonal matrix $D=\operatorname{diag}(4,-1)$. This means \begin{align*} T(v_1)=4v_1 \end{align*} and \begin{align*} T(v_2)=-v_2. \end{align*} Thus $v_1$ is an eigenvector with eigenvalue $4$, and $v_2$ is an eigenvector with eigenvalue $-1$. By the definition of matrix trace, the trace of $D$ is the sum of its diagonal entries: \begin{align*} \operatorname{tr}(D)=4+(-1)=3. \end{align*} Since $D$ represents $T$ in the chosen basis, the trace of the operator is \begin{align*} \operatorname{tr}(T)=\operatorname{tr}(D)=3. \end{align*} If another basis is chosen, the representing matrix has the form $P^{-1}DP$ for some invertible matrix $P$. By *[Trace Is Invariant Under Similarity](/theorems/7811)*, \begin{align*} \operatorname{tr}(P^{-1}DP)=\operatorname{tr}(D)=3. \end{align*} So the trace is still $3$ in every basis: the number belongs to the linear operator $T$, not to one particular coordinate display. [/example] ### Inseparable Extensions The separability hypothesis in the conjugate formula is not cosmetic. In positive characteristic, a finite extension can have too few embeddings, and trace can behave very differently from the separable picture. [example: Purely Inseparable Trace Can Vanish] Let $k=\mathbb{F}_p(t)$ and let $K=k(u)$ with $u^p=t$. The polynomial $X^p-t\in k[X]$ has derivative $pX^{p-1}=0$ in characteristic $p$, and $t$ is not a $p$th power in $\mathbb{F}_p(t)$, so $X^p-t$ is purely inseparable of degree $p$. Thus $(1,u,\dots,u^{p-1})$ is a $k$-basis of $K$. Let $e_j=u^{j-1}$ for $1\le j\le p$. Multiplication by $u$ sends \begin{align*} m_u(e_j)=u\cdot u^{j-1}=u^j=e_{j+1} \end{align*} for $1\le j\le p-1$, while for the last basis vector, \begin{align*} m_u(e_p)=u\cdot u^{p-1}=u^p=t=t e_1. \end{align*} Therefore the representing matrix $M$ of $m_u$ has $M_{j+1,j}=1$ for $1\le j\le p-1$, has $M_{1,p}=t$, and has every diagonal entry $M_{jj}=0$. By the definitions of field trace and matrix trace, \begin{align*} \operatorname{Tr}_{K/k}(u)=\operatorname{tr}(m_u)=\operatorname{tr}(M)=\sum_{j=1}^p M_{jj}=0+\cdots+0=0. \end{align*} The degeneracy is even stronger here. For $0\le r\le p-1$, multiplication by $u^r$ sends $e_j$ to a scalar multiple of $e_{j+r}$ with indices reduced using $u^p=t$; if $1\le r\le p-1$, no basis vector is sent to a scalar multiple of itself, so $\operatorname{Tr}_{K/k}(u^r)=0$. Also, \begin{align*} \operatorname{Tr}_{K/k}(1)=\operatorname{tr}(\operatorname{id}_K)=p=0 \end{align*} in characteristic $p$. Hence every $\alpha=\sum_{r=0}^{p-1} c_r u^r$ has trace $0$ by $k$-linearity of the matrix trace, and the trace pairing is identically zero in this purely inseparable example. [/example] ## Additivity and Cyclicity ### Linearity The first structural property is linearity. Trace forgets most of an endomorphism, but it preserves addition and scalar multiplication, which is why it functions as a linear functional on endomorphism spaces. [quotetheorem:7814] This theorem upgrades trace from a numerical invariant of one operator to a linear functional on the whole endomorphism space $\operatorname{End}_k(V)$. It justifies computations where matrices are decomposed into simpler pieces: diagonal parts, nilpotent parts, sums of projections, or scalar multiples can be traced term by term. That is why examples often compute trace after choosing a convenient basis or splitting an operator into an easier sum. Linearity also marks a boundary of what trace remembers. It preserves additive information but discards much of the multiplicative structure of an operator, so it cannot recover an endomorphism from its trace. The next structural property concerns multiplication more subtly: even though trace is not multiplicative in general, it is insensitive to cyclic reordering of compatible products. ### Cyclicity and Commutators The next theorem is needed because trace is also compatible with composition in a way that ordinary products are not. Products of maps may fail to commute, but trace can still cycle two compatible factors. This identity powers many manipulations in linear algebra and representation theory. [quotetheorem:401] The cyclic identity suggests that trace cannot see the order defect measured by a commutator. To make that statement precise, we first name the commutator operation on endomorphisms. [definition: Commutator of Endomorphisms] Let $k$ be a field and let $V$ be a $k$-vector space. The commutator of endomorphisms is the map \begin{align*} [\cdot,\cdot]: \operatorname{End}_k(V) \times \operatorname{End}_k(V) \to \operatorname{End}_k(V). \end{align*} It sends $(S,T)$ to \begin{align*} ST - TS. \end{align*} [/definition] Commutators measure failure of two maps to commute. Since trace is built from cyclic diagonal sums, the noncommuting parts of a product should cancel when arranged as $ST-TS$. The theorem below records this cancellation as a usable identity. It is the local fact that lets trace ignore commutator errors and later descend to abelianized constructions in algebra. [quotetheorem:7817] The commutator theorem says that trace ignores the part of a product that is visible only through noncommuting order. In practice this lets calculations replace $ST$ by $TS$ inside a trace expression, or show that an operator built as a commutator contributes no trace. This is one reason trace naturally appears on abelianized objects, where commutators are treated as negligible. The theorem also gives a warning about the size of the kernel of the trace functional: nonzero endomorphisms can have trace zero for structural reasons. Trace is therefore powerful for detecting certain averaged or first-order quantities, but it cannot distinguish every operator. Field trace has its own linearity property, developed next. ### Field Trace in Towers The field trace has a different-looking domain and codomain from ordinary endomorphism trace: it sends elements of $K$ back to scalars in $k$. To use it in linear algebra over the base field, one must know that addition in $K$ and scaling by elements of $k$ pass through the trace operation. This is a real compatibility condition, not just notation: an element of $K$ acts on $K$ by a $k$-linear multiplication operator, and the trace must turn sums and $k$-scalar multiples of such multiplication operators into the corresponding sums and scalar multiples in $k$. Without this linearity, trace pairings, dual bases, and computations with chosen bases of field extensions would not behave like ordinary linear algebra over the base field. The point to check is that the assignment $a \mapsto \operatorname{Tr}_{K/k}(a)$ is itself a $k$-linear map from the field extension $K$ to the base field $k$. The following result isolates exactly that property, so that field trace can be used as an ordinary linear functional in later arguments. [quotetheorem:7815] Linearity of field trace is what makes $\operatorname{Tr}_{K/k}:K\to k$ an ordinary $k$-linear functional, not just a rule attached to multiplication operators. It permits basis computations, lets trace pairings be bilinear, and supports dual-basis arguments where one solves linear equations using values of $\operatorname{Tr}_{K/k}(\alpha\beta)$. Without this theorem, the trace pairing would be a formula but not a reliable linear-algebraic tool. The scalar field matters here: the map is $k$-linear, not $K$-linear in general. Multiplying by an element outside $k$ changes the multiplication endomorphism in a way that trace does not simply factor out. The next compatibility question asks how this $k$-linear functional behaves when the extension is built through an intermediate field. Field extensions often occur in towers $L/K/k$, where an element can first be traced from $L$ to $K$ and then from $K$ to $k$. A trace operation that ignored this tower structure would be hard to compute consistently. The compatibility statement is that tracing in stages gives the same scalar as tracing directly to the base field. [quotetheorem:1293] Transitivity is the computational rule that makes trace usable in a tower one floor at a time. Instead of expanding $L$ directly as a $k$-vector space, one may first trace an element down to $K$ and then trace the result down to $k$; the theorem guarantees that no scalar information is lost or counted twice in this staged calculation. This is especially useful when a large extension is built from smaller ones whose traces are already known. The theorem also records the hypothesis hidden in this kind of computation: all extensions in the tower must be finite so that the relevant multiplication maps have finite traces. It does not say that the trace pairing is nondegenerate, nor does it remove inseparability phenomena. Those issues belong to the next theorem, where trace is used not merely as a scalar-valued map but as a test for separability. The trace pairing can fail to distinguish elements if some nonzero element pairs to zero with everything else. That degeneracy is not just a defect of the chosen basis; in finite field extensions it reflects inseparability. Nondegeneracy of the trace form is therefore the linear-algebraic shadow of separability. [quotetheorem:7816] This theorem turns the trace pairing into a separability detector. For a separable finite extension, every nonzero element can be tested against some partner so that the trace of the product is nonzero; equivalently, the trace form has no radical. That is why discriminants built from trace matrices vanish exactly when the chosen basis fails to witness separability in the expected way. The converse behavior explains the earlier purely inseparable example: the trace pairing may collapse even when the extension has positive degree and a perfectly good vector-space basis. Nondegeneracy is therefore not a formal consequence of bilinearity or symmetry; it is a genuine arithmetic property of the extension. ## Relationship to Other Concepts ### Trace Forms and Characters Trace sits at the intersection of additive and multiplicative algebra. As an additive invariant, it is linear in the operator. As part of the characteristic polynomial, it is tied to determinant, eigenvalues, and the [Cayley-Hamilton theorem](/theorems/865). As a field-theoretic operation, it supports discriminants, dual bases, and arithmetic of number fields. The trace pairing becomes especially useful when viewed as a bilinear form. Naming the trace form makes it easier to connect trace to discriminants and to the linear algebra of dual bases. [definition: Trace Form] Let $K/k$ be a finite field extension. The trace form is the map \begin{align*} \operatorname{TrForm}_{K/k}: K \times K \to k. \end{align*} It sends $(\alpha,\beta)$ to \begin{align*} \operatorname{Tr}_{K/k}(\alpha\beta). \end{align*} [/definition] The trace form produces concrete scalar invariants once a basis is chosen. A basis gives a Gram matrix whose $(i,j)$ entry records the trace pairing of the $i$th and $j$th basis elements. Taking its determinant packages whether the trace form is degenerate on that basis and creates the discriminant used in arithmetic applications. [definition: Discriminant of a Basis] Let $K/k$ be a finite field extension of degree $n$. The discriminant of an ordered $k$-basis is the map \begin{align*} \operatorname{disc}: \{\text{ordered }k\text{-bases of }K\} \to k. \end{align*} It sends $(\alpha_1,\dots,\alpha_n)$ to \begin{align*} \det\big(\operatorname{Tr}_{K/k}(\alpha_i\alpha_j)\big)_{1 \le i,j \le n}. \end{align*} [/definition] Trace also appears in representation theory, where each group element acts by a linear map after a representation is chosen. Individual matrix entries depend on the basis of the representation space, but trace survives conjugation of the representing matrices. This motivates packaging those traces into a single function on the group. The definition below names that function, which is the standard character attached to a finite-dimensional representation. [definition: Character of a Finite-Dimensional Representation] Let $G$ be a group, let $k$ be a field, let $V$ be a finite-dimensional $k$-vector space, and let $GL(V)$ denote the group of invertible $k$-linear maps from $V$ to itself. Let $\rho: G \to GL(V)$ be a representation. The character of $\rho$ is the function \begin{align*} \chi_\rho: G \to k. \end{align*} It sends $g$ to \begin{align*} \operatorname{tr}(\rho(g)). \end{align*} [/definition] [Characters are constant on conjugacy classes](/theorems/5008) because trace is invariant under similarity. This is one of the main reasons trace is the scalar invariant used in finite group representation theory, rather than determinant or a chosen matrix entry. ### Boundary Cases [remark: Trace Versus Determinant] The trace and [determinant](/page/Determinant) are complementary invariants of an endomorphism. Trace is additive and detects the coefficient of $t^{n-1}$ in the characteristic polynomial, while determinant is multiplicative and gives the constant term up to sign. [/remark] This complementarity has a boundary: both invariants are simplest when the underlying linear object is finite-dimensional. Trace in particular is not just a formal diagonal sum once infinitely many diagonal entries are present. [remark: Trace Depends on Finiteness] For finite-dimensional vector spaces and finite free modules, trace is a finite diagonal sum. Infinite-dimensional settings require extra hypotheses, such as trace-class operators, before a meaningful trace can be defined. [/remark] ## Beyond and Connected Topics Trace is one entry point into a larger package of invariants. The [determinant](/page/Determinant) records multiplicative volume change, while trace records the additive first-order coefficient of the characteristic polynomial. Together with the remaining characteristic-polynomial coefficients, they connect [matrix](/page/Matrix) computations to [eigenvalues and eigenvectors](/page/Eigenvalue%20and%20Eigenvector), Cayley-Hamilton theory, and representation characters. In [field extension](/page/Field%20Extension) theory, the trace pairing leads directly to discriminants, dual bases, and separability criteria. Algebraic number theory uses these ideas to measure integral bases and ramification, while Galois theory interprets trace through sums over embeddings. In representation theory, the same basis-invariance principle makes characters class functions and turns linear algebra into a tool for studying group structure. ## References [Linear Map](/page/Linear%20Map). [Matrix](/page/Matrix). [Field Extension](/page/Field%20Extension). Lang, *Algebra* (2002). Dummit and Foote, *Abstract Algebra* (2004). Atiyah and Macdonald, *Introduction to Commutative Algebra* (1969).