The Killing Form, Its Radical and the Centre of $\mathfrak{gl}_n(\mathbb{R})$
Introduction
The Killing form of $\mathfrak{gl}_n(\mathbb{R})$ is $\operatorname{K}(X,Y)=2n\operatorname{tr}(XY)-2\operatorname{tr}(X)\operatorname{tr}(Y)$, its radical is the centre $\mathbb{R}\mathrm{I}$, and on the quotient $\mathfrak{sl}_n(\mathbb{R})$ it is $2n$ times the trace form, of inertia $\bigl(n(n+1)/2-1,\,n(n-1)/2\bigr)$.
The article builds the adjoint representation in the $n^{2}$-dimensional basis of matrix units, computes the form and checks it numerically at $n=2,3,4$, identifies the radical, and computes the inertia on the quotient and on $\mathfrak{so}(n)$. It assumes The Trace Form and Its Indefinite Inertia for the trace form and The Hilbert–Schmidt Form: Definite but Not Invariant for the definite one, and it defers the comparison of the four forms to The Four Forms and Their Algebraic Norms over $\mathbb{R}$.
The Adjoint Representation
The adjoint map $\operatorname{ad}_X(Y)=[X,Y]$ is a derivation of the algebra. It is linear in $Y$, and the Jacobi identity gives
$$ \operatorname{ad}_X(YZ)=[X,YZ]=[X,Y]Z+Y[X,Z]=\operatorname{ad}_X(Y)\,Z+Y\,\operatorname{ad}_X(Z). $$
So $\operatorname{ad}_X$ is an endomorphism of $M_n(\mathbb{R})$ as an algebra, and it is a linear endomorphism of the $n^{2}$-dimensional vector space.
Define $\operatorname{K}(X,Y)=\operatorname{tr}(\operatorname{ad}_X\operatorname{ad}_Y)$. The trace is taken on the $n^{2}$-dimensional space. The form is bilinear and symmetric, because the trace of a product of two endomorphisms is symmetric in the factors.
The form is invariant under the adjoint action. For all $X,Y,Z$,
$$ \operatorname{K}([Z,X],Y)+\operatorname{K}(X,[Z,Y])=0, $$
because the left-hand side is $\operatorname{tr}\bigl([\operatorname{ad}_Z,\operatorname{ad}_X]\operatorname{ad}_Y+\operatorname{ad}_X[\operatorname{ad}_Z,\operatorname{ad}_Y]\bigr)$ and the trace of a commutator vanishes. So the form is a symmetric invariant bilinear form on the Lie algebra, and it is also invariant under $\operatorname{Ad}_S$ for $S\in\mathrm{GL}_n(\mathbb{R})$.
The Computation
The form is $\operatorname{K}(X,Y)=2n\operatorname{tr}(XY)-2\operatorname{tr}(X)\operatorname{tr}(Y)$. The proof decomposes the adjoint map into a left and a right multiplication.
Proof. Let $L_X(Y)=XY$ and $R_X(Y)=YX$. Then $\operatorname{ad}_X=L_X-R_X$, the two families commute, and $L_XL_Y=L_{XY}$ while $R_XR_Y=R_{YX}$. On the matrix units,
$$ L_XL_Y(E_{ij})=XYE_{ij},\qquad R_XR_Y(E_{ij})=E_{ij}YX,\qquad L_XR_Y(E_{ij})=XE_{ij}Y. $$
The coefficient of $E_{ij}$ in $XYE_{ij}$ is $(XY)_{ii}$, so $\operatorname{tr}(L_XL_Y)=n\operatorname{tr}(XY)$. The coefficient of $E_{ij}$ in $E_{ij}YX$ is $(YX)_{jj}$, so $\operatorname{tr}(R_XR_Y)=n\operatorname{tr}(YX)=n\operatorname{tr}(XY)$. The coefficient of $E_{ij}$ in $XE_{ij}Y$ is $X_{ii}Y_{jj}$, so $\operatorname{tr}(L_XR_Y)=\operatorname{tr}(X)\operatorname{tr}(Y)$. Hence
$$ \operatorname{K}(X,Y)=\operatorname{tr}\bigl((L_X-R_X)(L_Y-R_Y)\bigr)=2n\operatorname{tr}(XY)-2\operatorname{tr}(X)\operatorname{tr}(Y). $$
The formula was checked numerically. The adjoint map was built as an $n^{2}\times n^{2}$ matrix in the matrix-unit basis, the trace of the product of two such matrices was compared with the formula, and the comparison was made on a hundred random pairs at $n=2$, $n=3$ and $n=4$. The maximal absolute deviation was $3.6\times10^{-15}$ at $n=2$ and $1.4\times10^{-14}$ at $n=3$ and $n=4$, on values of $\operatorname{K}$ reaching $16.3$, $59.2$ and $86.9$.
The formula separates the centre from the rest. The second term depends only on the traces, so it vanishes on the trace-zero subspace. On $\mathfrak{sl}_n(\mathbb{R})$ the form is $2n$ times the trace form.
The Radical
The radical of the form is the centre. Let $X$ satisfy $\operatorname{K}(X,Y)=0$ for all $Y$.
Step one: the off-diagonal entries vanish. Take $Y=E_{ij}$ with $i\ne j$. Then $\operatorname{tr}(XY)=X_{ji}$ and $\operatorname{tr}(Y)=0$, so $2nX_{ji}=0$ and $X_{ji}=0$.
Step two: the diagonal entries are equal. Take $Y=E_{ii}$. Then $\operatorname{tr}(XY)=X_{ii}$ and $\operatorname{tr}(Y)=1$, so $2nX_{ii}-2\operatorname{tr}(X)=0$, which gives $X_{ii}=\operatorname{tr}(X)/n$ for every $i$. All diagonal entries have the same value.
Conclusion. $X=\lambda\mathrm{I}$ with $\lambda=\operatorname{tr}(X)/n$. Conversely every multiple of the identity is in the radical, because $\operatorname{K}(\mathrm{I},Y)=2n\operatorname{tr}(Y)-2n\operatorname{tr}(Y)=0$ for every $Y$. Hence
$$ \operatorname{rad}\operatorname{K}=\mathbb{R}\mathrm{I}=\mathfrak{z}\bigl(\mathfrak{gl}_n(\mathbb{R})\bigr), $$
a space of dimension one. The radical of the form is exactly the centre of the algebra.
The numerical check agrees. The matrix of the form in the matrix-unit basis has nullity one at $n=2,3,4$, and the null vector is proportional to the identity each time.
The inertia counts the radical separately. The matrix of $\operatorname{K}$ in the matrix-unit basis has inertia $(2,1,1)$ at $n=2$, $(5,3,1)$ at $n=3$ and $(9,6,1)$ at $n=4$. The last number is the dimension of the radical.
The Inertia on the Quotient
On $\mathfrak{sl}_n(\mathbb{R})$ the form is twice $n$ times the trace form. For $X,Y$ of trace zero, the second term of the formula vanishes, so $\operatorname{K}(X,Y)=2n\operatorname{tr}(XY)$. The factor $2n>0$ does not change signs, so the inertia of $\operatorname{K}$ on $\mathfrak{sl}_n(\mathbb{R})$ is the inertia of the trace form there.
The trace form on $\mathfrak{sl}_n(\mathbb{R})$ splits into a positive and a negative part. The trace-zero subspace meets $\operatorname{Sym}_n$ in the symmetric matrices of trace zero, of dimension $n(n+1)/2-1$, and it meets $\operatorname{Skew}_n$ in the whole of $\operatorname{Skew}_n$, of dimension $n(n-1)/2$. The two parts are orthogonal, and they exhaust the subspace because the dimensions add to $n^{2}-1$. The form is positive on the first and negative on the second.
The inertia is $\bigl(n(n+1)/2-1,\,n(n-1)/2\bigr)$. The values are $(2,1)$ at $n=2$, $(5,3)$ at $n=3$ and $(9,6)$ at $n=4$, and they were measured on the matrix of the restricted form in an explicit basis of $\mathfrak{sl}_n(\mathbb{R})$: the off-diagonal units together with the differences $E_{ii}-E_{nn}$.
The Restriction to $\mathfrak{so}(n)$
On $\mathfrak{so}(n)$ the form is a negative multiple of the trace form. The elements of $\mathfrak{so}(n)$ are the antisymmetric matrices, and they have trace zero, so $\operatorname{K}(X,Y)=2n\operatorname{tr}(XY)$ there. For antisymmetric $X$ the trace of $X^{2}$ is $-2\sum_{i The eigenvalues in the standard basis are negative and explicit. Take the basis $L_{ij}=E_{ij}-E_{ji}$ for $i The intrinsic Killing form of $\mathfrak{so}(n)$ is $\operatorname{K}_{\mathfrak{so}}(X,Y)=(n-2)\operatorname{tr}(XY)$. It is computed from the adjoint representation of $\mathfrak{so}(n)$ itself, of dimension $n(n-1)/2$, and not from the adjoint representation of $\mathfrak{gl}_n(\mathbb{R})$. On the basis $L_{ij}$ its eigenvalues are $-2(n-2)$: that is $-2$ three times at $n=3$ and $-4$ six times at $n=4$, as measured. The two forms differ by the factor $2n$, which is the expected ratio, since both are invariant forms on a simple algebra. At $n=2$ the restriction vanishes identically. The algebra $\mathfrak{so}(2)$ is one-dimensional and abelian, spanned by $J=E_{12}-E_{21}$. Its adjoint representation is zero, so $\operatorname{K}(J,J)=\operatorname{tr}(0)=0$, and the formula gives $4\operatorname{tr}(J^{2})=-8$ for the ambient form but $0$ for the intrinsic one. An abelian algebra has a zero Killing form, and the two-dimensional case is the first instance. A Lie algebra is semisimple exactly when its Killing form is nondegenerate. This is the Cartan criterion, and it applies to each of the three algebras here. $\mathfrak{gl}_n(\mathbb{R})$ is reductive and not semisimple. Its form has a radical of dimension one, the centre, so it is degenerate and the algebra is not semisimple. It is the direct sum $\mathbb{R}\mathrm{I}\oplus\mathfrak{sl}_n(\mathbb{R})$ of Lie algebras, because the identity is central, and each summand is either abelian or semisimple. That is the definition of reductive. $\mathfrak{sl}_n(\mathbb{R})$ is semisimple. The restricted form has inertia $\bigl(n(n+1)/2-1,n(n-1)/2\bigr)$, with no zero direction, so it is nondegenerate. The centre of $\mathfrak{sl}_n(\mathbb{R})$ is zero for $n\ge2$, and the criterion applies. $\mathfrak{so}(n)$ is semisimple for $n\ge3$. The intrinsic Killing form is $(n-2)\operatorname{tr}(XY)$, which is definite, so it is nondegenerate for $n\ge3$, and the factor $n-2$ is nonzero. At $n=2$ the algebra is abelian and the criterion fails: $\mathfrak{so}(2)$ is not semisimple. The centre is the radical of the form, and this is the whole content of the criterion here. The three statements above are one computation read three ways: the radical of $\operatorname{K}$ is $\mathbb{R}\mathrm{I}$, the quotient by it is $\mathfrak{sl}_n(\mathbb{R})$, and the form is nondegenerate there. The two forms agree up to a factor on the trace-zero part. On $\mathfrak{sl}_n(\mathbb{R})$ the Killing form is $2n$ times the trace form. So the inertia of the Killing form is the inertia of the trace form minus one positive direction, the direction of the identity. The difference is the correction that kills the identity. The term $-2\operatorname{tr}(X)\operatorname{tr}(Y)$ is what makes $\operatorname{K}(\mathrm{I},\mathrm{I})$ vanish. Without it the form would be $2n\operatorname{tr}(XY)$, which is the trace form rescaled, of inertia $\bigl(n(n+1)/2,n(n-1)/2\bigr)$ and nondegenerate. The Killing form is that form made degenerate along the centre. The Killing form of $\mathfrak{gl}_n(\mathbb{R})$ is $\operatorname{K}(X,Y)=2n\operatorname{tr}(XY)-2\operatorname{tr}(X)\operatorname{tr}(Y)$, computed by writing $\operatorname{ad}_X=L_X-R_X$ and taking the traces of the four products of the left and right multiplications. The formula was verified on a hundred random pairs at $n=2,3,4$ with a maximal deviation of $1.4\times10^{-14}$. The radical is the centre $\mathbb{R}\mathrm{I}$, of dimension one, and it is exactly the set of elements orthogonal to the whole algebra. On $\mathfrak{sl}_n(\mathbb{R})$ the form is $2n$ times the trace form, of inertia $\bigl(n(n+1)/2-1,n(n-1)/2\bigr)$, measured as $(2,1)$, $(5,3)$ and $(9,6)$ at $n=2,3,4$. On $\mathfrak{so}(n)$ the restriction is negative definite: in the basis $L_{ij}=E_{ij}-E_{ji}$ the trace form has the eigenvalue $-2$ with multiplicity $n(n-1)/2$, the restricted form has $-4n$, and the intrinsic Killing form of $\mathfrak{so}(n)$, which is $(n-2)\operatorname{tr}(XY)$, has $-2$ three times at $n=3$ and $-4$ six times at $n=4$. At $n=2$ the restriction vanishes, because $\mathfrak{so}(2)$ is abelian. By the Cartan criterion $\mathfrak{gl}_n(\mathbb{R})$ is reductive and not semisimple, while $\mathfrak{sl}_n(\mathbb{R})$ and $\mathfrak{so}(n)$ for $n\ge3$ are semisimple.The Cartan Criterion
The Form and the Trace Form
Summary
Summary of Notation
Symbol
Meaning
$\operatorname{ad}_X(Y)=[X,Y]$
the adjoint map, a derivation
$\operatorname{K}(X,Y)=\operatorname{tr}(\operatorname{ad}_X\operatorname{ad}_Y)$
the Killing form
$L_X,R_X$
left multiplication by $X$ and right multiplication by $X$
$E_{ij}$
the matrix unit, $1$ in the entry $(i,j)$
$\mathrm{I}$
the identity matrix
$\mathfrak{gl}_n(\mathbb{R})$
the algebra $M_n(\mathbb{R})$ read as a Lie algebra
$\mathfrak{sl}_n(\mathbb{R})$
the trace-zero subspace, of dimension $n^{2}-1$
$\mathfrak{so}(n)$
the antisymmetric matrices, of dimension $n(n-1)/2$
$L_{ij}=E_{ij}-E_{ji}$
the standard basis of $\mathfrak{so}(n)$
$\operatorname{rad}\operatorname{K}$
the radical of the form, equal to $\mathbb{R}\mathrm{I}$
$\mathfrak{z}(\mathfrak{gl}_n(\mathbb{R}))$
the centre, equal to $\mathbb{R}\mathrm{I}$
Further Reading