The Elementary Symmetric Functions and $\det(\mathrm{I}+tA)$ at $n=4$
Introduction
The four coefficients of $\det(\mathrm I+tA)$ are the elementary symmetric functions of the eigenvalues, and they are polynomials in the traces of the powers of $A$: $e_1=\operatorname{tr}A$ up to the trace itself, $e_4=\det A$, and the two middle ones carry the information the ends lose.
The article defines the four coefficients, proves Newton's identities that express them through the power sums, identifies $e_k$ with the trace of $A$ on the $k$-th exterior power so that the ladder of dimensions $1,4,6,4,1$ appears, shows that each $e_k$ is invariant under conjugation and that each level set $\{e_k=0\}$ is a hypersurface, and reads the four coefficients in the axis-free basis, where $e_1=4c_{tt}$ and $e_4$ is the square of the quaternionic norm.
It assumes The Element, the Coordinates and the Trace of $M_4(\mathbb{C})$ for the coordinates, and The Matrix Algebra $M_4(\mathbb{C})$ for the product and the determinant. The general order is The Elementary Symmetric Functions, the Trace and the Determinant; the case $n=2$ is The Elementary Symmetric Functions: $\operatorname{tr}$ and $\det$ in Dimension Two. The spectral reading of the same numbers is The Spectral Theorem: the Hermitian Case at $n=4$.
The Four Coefficients
Definition. For $A\in M_4(\mathbb{C})$ the elementary symmetric functions are the coefficients of the polynomial in $t$,
$$ \det(\mathrm I+tA)=\sum_{k=0}^{4}e_k(A)\,t^{k},\qquad e_0=1,\quad e_4=\det A . $$
Proposition. The coefficient $e_k$ is the $k$-th elementary symmetric function of the eigenvalues, counted with multiplicity:
$$
e_k=\sum_{i_1<\cdots Proof. Over $\mathbb{C}$ the matrix is triangularisable, $A=PTP^{-1}$ with $T$ upper triangular; then $\mathrm I+tA=P(\mathrm I+tT)P^{-1}$ and $\det(\mathrm I+tT)=\prod_i(1+t\lambda_i)$. Expanding the product collects $t^{k}$ with the sum of the $k$-fold products of distinct eigenvalues. Corollary. The four coefficients are $e_1=\sum\lambda_i$, $e_2=\sum_{i Proof. Expand the product $\prod_i(1+t\lambda_i)$ and read the coefficient of $t^{k}$. The polynomial at $t=-1$ and at large $t$. At $t=-1$ the identity gives $\det(\mathrm I-A)=\sum_k(-1)^{k}e_k$, the value at the eigenvalue $1$; at $t\to\infty$ it gives $e_4=\det A$. The four coefficients thus interpolate between the trace and the determinant. Definition. The power sums are $p_k=\operatorname{tr}(A^{k})=\sum_i\lambda_i^{k}$, for $k=1,2,3,4$. Theorem (Newton). For $1\le k\le4$, $$
k\,e_k=\sum_{i=1}^{k}(-1)^{i-1}e_{k-i}\,p_i .
$$ Proof. Differentiate the identity $\log\det(\mathrm I+tA)=\sum_i\log(1+t\lambda_i)$ at $t$ and multiply by $t$: the left side is $t\operatorname{tr}\bigl(A(\mathrm I+tA)^{-1}\bigr)$ and the right side is $\sum_{k\ge1}(-1)^{k-1}p_kt^{k}$. Comparing coefficients in the expansion of $E(t)=\det(\mathrm I+tA)$ gives the recursion $k e_k=\sum_{i=1}^{k}(-1)^{i-1}e_{k-i}p_i$. Corollary (the four identities written out). $$
p_1=e_1,\qquad p_2=e_1p_1-2e_2,\qquad p_3=e_1p_2-e_2p_1+3e_3,\qquad p_4=e_1p_3-e_2p_2+e_3p_1-4e_4 .
$$ Proof. Substitute $k=1,2,3,4$ in the theorem and collect. Corollary (the coefficients through the traces). $$
e_1=\operatorname{tr}A,\qquad
e_2=\tfrac12\bigl[(\operatorname{tr}A)^{2}-\operatorname{tr}(A^{2})\bigr],
$$ $$
e_3=\tfrac16\bigl[(\operatorname{tr}A)^{3}-3\operatorname{tr}A\operatorname{tr}(A^{2})+2\operatorname{tr}(A^{3})\bigr],\qquad
e_4=\det A .
$$ Proof. Solve the four identities for $e_1,e_2,e_3$ in turn: $e_2=\tfrac12(p_1^{2}-p_2)$, then $e_3=\tfrac13(p_3-e_1p_2+e_2p_1)$, which reduces to the display. The first three are determined by traces, the fourth is not. The traces of $A,A^{2},A^{3}$ determine $e_1,e_2,e_3$; the determinant needs one more number. Equivalently, $\mathrm{tr}(A^{k})$ for $k=1,2,3$ fixes the characteristic polynomial up to the constant term, and the constant term is $\det A$. Theorem. $e_k(A)=\operatorname{tr}\bigl(\Lambda^{k}A\bigr)$, the trace of the induced map on the $k$-th exterior power of $\mathbb{C}^{4}$. Proof. The eigenvalues of $\Lambda^{k}A$ are the products $\lambda_{i_1}\cdots\lambda_{i_k}$ over the $k$-element subsets, each once; their sum is $e_k$ by the first proposition. Corollary. The four coefficients live on spaces of dimensions $1,4,6,4,1$: $$
\dim\Lambda^{0}=\dim\Lambda^{4}=1,\qquad \dim\Lambda^{1}=\dim\Lambda^{3}=4,\qquad \dim\Lambda^{2}=6 .
$$ Proof. The binomial coefficients $\binom{4}{k}$ for $k=0,1,2,3,4$, and $\Lambda^{2}\mathbb{C}^{4}$ also has dimension the dimension of the antisymmetric half, $6$. The two readings of the middle coefficient. $e_2$ is the trace on $\Lambda^{2}\mathbb{C}^{4}$, and it is also the sum of the six principal $2\times2$ minors of $A$. The identity $\operatorname{tr}(\Lambda^{2}A)=\sum_{i The four coefficients and the four forms. $e_1$ and $e_4$ are the trace and the determinant of the regular representation; $e_2$ and $e_3$ are the coefficients that distinguish the two middle exterior powers. At $n=4$ all four are needed: the characteristic polynomial has degree four and no two of its coefficients are functions of the others. Proposition. Each $e_k$ is invariant under conjugation: $e_k(PAP^{-1})=e_k(A)$ for invertible $P$. Proof. The eigenvalues are unchanged by conjugation, and each $e_k$ is a symmetric function of them. Equivalently, $\det(\mathrm I+tPAP^{-1})=\det P\det(\mathrm I+tA)\det P^{-1}$. Corollary. Each $e_k$ is a polynomial in the entries invariant under $\mathrm{GL}_4$, and the four together generate the invariant ring of the conjugation action on the sixteen coordinates. Proof. The invariant ring of $n\times n$ matrices under conjugation is generated by the coefficients of the characteristic polynomial, and at $n=4$ those are $e_1,e_2,e_3,e_4$. The strata. Each set $\{e_k=0\}$ is a hypersurface, of complex dimension $15$ in the space of dimension $16$, except that $\{e_2=0\}$ and $\{e_3=0\}$ are singular at the origin. Proposition. The singular set $\{e_4=0\}$ is exactly the set of non-invertible matrices, and it is a hypersurface whose smooth part has complex dimension $15$. Proof. $\det A=0$ is the condition of non-invertibility, and the derivative of the determinant at $A$ is $X\mapsto\operatorname{tr}(\operatorname{adj}(A)X)$, which is nonzero whenever $\operatorname{adj}(A)\neq0$, that is whenever $\operatorname{rank}A\ge3$. The four coordinates. Write $A=\sum_{\mu,\nu}c_{\mu\nu}T_{\mu\nu}$ in the axis-free basis $T_{\mu\nu}=L_{\mathrm{axf},e_\mu}R_{\mathrm{axf},e_\nu}$; the matrix units are the basis in which the product is defined, and the axis-free sixteen are the basis in which the coefficients are read by traces. The trace is one coordinate, $e_1=\operatorname{tr}A=4c_{tt}$, and the determinant is the fourth coefficient. The left matrix. For $A=L_{\mathrm{axf}}(\tilde Q)$ the eigenvalues are $Q_t\pm iB$, each twice, with $B^{2}=Q_x^{2}+Q_y^{2}+Q_z^{2}$. Hence $$
\det(\mathrm I+tL_{\mathrm{axf}}(\tilde Q))=\bigl(1+2Q_tt+(Q_t^{2}+B^{2})t^{2}\bigr)^{2},
$$ and the four coefficients are $$
e_1=4Q_t,\qquad e_2=6Q_t^{2}+2B^{2},\qquad e_3=4Q_t(Q_t^{2}+B^{2}),\qquad e_4=(Q_t^{2}+B^{2})^{2}.
$$ Proof. The eigenvalues come in two conjugate pairs, so the characteristic polynomial is the square of a quadratic; expanding the square gives the display, and $e_1=4Q_t$ agrees with the trace, while $e_4=\det L_{\mathrm{axf}}(\tilde Q)=\langle\tilde Q\rangle_{\natural}^{2}$. The check at the identity and at the vector part. At $Q_t=1$ and $B=0$ the display gives $e_2=6$, $e_3=4$, $e_4=1$, which is $\det(\mathrm I+t\mathrm I_4)=(1+t)^{4}$. At $Q_t=0$ the display gives $e_1=e_3=0$ and $e_2=2B^{2}$, $e_4=B^{4}$: the traceless left matrices have the spectrum $\pm iB$ twice. What the axis-free reading adds. The determinant of the matrix is the square of a quadratic form, and the quadratic form is the quaternionic norm $\langle\tilde Q\rangle_{\natural}$, which is the sum of four squares over $\mathbb{C}$ and a Euclidean norm on the real slice. The other three coefficients are the same norm together with the scalar coordinate. In the matrix-unit basis the four coefficients are the same numbers, written as sums over $4^{k}$ minors; the axis-free basis writes them as sums over four coordinates. The four coefficients of $\det(\mathrm I+tA)$ are the elementary symmetric functions of the eigenvalues, and Newton's identities express the first three through $\operatorname{tr}A$, $\operatorname{tr}(A^{2})$ and $\operatorname{tr}(A^{3})$, with $e_4=\det A$ independent of them. The coefficient $e_k$ is the trace of $A$ on the $k$-th exterior power, so the four live on spaces of dimensions $1,4,6,4,1$, and the middle one is also the sum of the six principal minors. Each $e_k$ is invariant under conjugation, the four generate the invariant ring, and each level set is a hypersurface with the singular set $\{\det=0\}$ for the fourth. In the axis-free basis $e_1=4c_{tt}$ and $e_4=\langle\tilde Q\rangle_{\natural}^{2}$, and for a one-sided matrix the four coefficients are $4Q_t$, $6Q_t^{2}+2B^{2}$, $4Q_t(Q_t^{2}+B^{2})$ and $(Q_t^{2}+B^{2})^{2}$.Newton's Identities
The Exterior Powers and the Ladder $1,4,6,4,1$
Invariance and the Strata
Stratum
Equation
Reading
traceless hyperplane
$e_1=0$
$\mathfrak{sl}_4$, dimension $15$
the quadric
$e_2=0$
rank at most two, or trace data of a degenerate pair
the cubic
$e_3=0$
one eigenvalue zero, or three summing to zero
the singular set
$e_4=0$
$\det A=0$, rank at most three
The Axis-Free Reading
Summary
Summary of Notation
Symbol
Meaning
$e_k$
the $k$-th coefficient of $\det(\mathrm I+tA)$
$p_k$
$\operatorname{tr}(A^{k})$, the $k$-th power sum
$\lambda_i$
the four eigenvalues, with multiplicity
$\Lambda^{k}A$
the induced map on the $k$-th exterior power
$c_{\mu\nu}$
the axis-free coordinates, $c_{tt}=e_1/4$
$Q_t$, $B^{2}$
the scalar coordinate and $Q_x^{2}+Q_y^{2}+Q_z^{2}$
$\langle\tilde Q\rangle_{\natural}$
$Q_t^{2}+B^{2}$, whose square is $e_4$
Further Reading