The Exponential and the Lie Group Structure

Introduction

The group $\mathrm{GL}_n(\mathbb{C})$ with the commutator is a Lie group whose Lie algebra is $\mathfrak{gl}_n$, the exponential is a local diffeomorphism at the origin and a surjection onto the group over $\mathbb{C}$, and the one-parameter subgroups are exactly the exponentials $t\mapsto e^{tX}$.

The article proves that the commutator makes the matrix space a Lie algebra and that the group is a Lie group with that algebra, proves that the exponential is a local diffeomorphism near zero with the inverse given by the logarithm, proves that over $\mathbb{C}$ the exponential is surjective but never injective, proves that every continuous one-parameter subgroup is an exponential, and proves the first two terms of the Campbell–Baker–Hausdorff formula together with the identity $\operatorname{Ad}_{e^{X}}=e^{\operatorname{ad}_{X}}$. It assumes The Analysis on $M_n(\mathbb{C})$ and The Exponential of a Matrix and Its Image. It defers the two halves to The Two Halves and the Analysis: Smooth Structures and the Exponential and the decomposition to The Cartan Decomposition of $\mathrm{GL}_n(\mathbb{C})$.

The Lie Algebra

The commutator makes the space a Lie algebra. With

$$ [X,Y]=XY-YX $$

the space is a Lie algebra, since the bracket is bilinear, antisymmetric and satisfies the Jacobi identity

$$ [X,[Y,Z]]+[Y,[Z,X]]+[Z,[X,Y]]=0. $$

The computation confirms the Jacobi identity at $n=3$ on random triples, and it is the algebra whose exponential gives the group.

The derived algebra is the traceless part. Since $\operatorname{tr}[X,Y]=0$,

$$ [\mathfrak{gl}_n,\mathfrak{gl}_n]=\mathfrak{sl}_n=\{X:\operatorname{tr}X=0\}, $$

the traceless matrices, of complex dimension $n^{2}-1$. The computation confirms the vanishing of the trace at $n=3$ on random commutators and the dimension $8$ of $\mathfrak{sl}_3$, and it is the first structural fact of the algebra.

The centre is the scalars. An element commuting with every matrix is a scalar, so

$$ \mathfrak{z}(\mathfrak{gl}_n)=\mathbb{C}\,\mathrm{I}, $$

and the algebra is the direct sum of the scalar line and the traceless part, both ideals. The computation confirms the centre at $n=3$ by commuting a random matrix with the whole basis, and it gives the split

$$ \mathfrak{gl}_n=\mathfrak{sl}_n\oplus\mathbb{C}\mathrm{I},\qquad \dim_{\mathbb{C}}=n^{2}. $$

The group is a Lie group with this algebra. The group is an open submanifold of the space of real dimension $2n^{2}$, the multiplication and the inversion are smooth, and the tangent space at the identity is $\mathfrak{gl}_n$; so

$$ \operatorname{Lie}\big(\mathrm{GL}_n(\mathbb{C})\big)=\mathfrak{gl}_n,\qquad \dim_{\mathbb{R}}\mathrm{GL}_n(\mathbb{C})=2n^{2}. $$

The computation confirms the dimensions $8,18,32$ at $n=2,3,4$, and it is the reason the group and its algebra are the two levels of the same structure.

The Exponential as a Local Diffeomorphism

The exponential is smooth with the identity as its derivative at zero. From the analysis of the space,

$$ D\exp(0)\,H=H, $$

so the exponential is a local diffeomorphism near the origin by the inverse function theorem. The computation confirms the derivative at $n=3$ by a central difference, and it gives a chart of the group near the identity.

The logarithm is the local inverse. For $\|A-\mathrm{I}\|<1$ the series

$$ \log A=\sum_{k\ge1}\frac{(-1)^{k+1}}{k}(A-\mathrm{I})^{k} $$

converges and satisfies $\exp(\log A)=A$ and $\log(\exp X)=X$ for $\|X\|$ small. The computation confirms the two identities at $n=3$ on random matrices near the identity and random matrices of small norm, and it is the reason the exponential is a chart.

The derivative of the exponential is invertible exactly off the resonance set. The kernel of $D\exp(X)$ is nontrivial exactly when two eigenvalues of $X$ differ by a nonzero multiple of $2\pi i$, so

$$ D\exp(X)\ \text{is invertible}\iff \lambda_i-\lambda_j\notin 2\pi i\mathbb{Z}\setminus\{0\}. $$

The computation confirms the statement at $n=2$ on the matrix $\operatorname{diag}(0,2\pi i)$, where the derivative is singular, and on a matrix with a generic spectrum, where it is invertible.

The Global Behaviour of the Exponential

Over $\mathbb{C}$ the exponential is surjective. Every invertible complex matrix has a logarithm: by the Jordan form it is enough to take the logarithm of a Jordan block with a nonzero eigenvalue, and the series or the integral gives it. Hence

$$ \exp\big(\mathfrak{gl}_n\big)=\mathrm{GL}_n(\mathbb{C}). $$

The computation confirms the statement at $n=3$ by computing the logarithm of a random invertible matrix and exponentiating back, and it is the place where the complex case differs from the real one.

The exponential is never injective. Since the scalar $2\pi i\mathrm{I}$ is in the kernel,

$$ e^{X+2\pi i k\mathrm{I}}=e^{X}\qquad\text{for every integer }k, $$

so the exponential has an infinite discrete fibre over every point of the group. The computation confirms the identity at $n=3$ on a random matrix and the integer $k=1$, and it is the reason the logarithm is defined only up to the periods $2\pi i\mathbb{Z}$.

The exponential is not a covering of the whole group. The fibre over a matrix is a discrete set, but the number of sheets is not constant, since the periods depend on the multiplicities of the eigenvalues; so

$$ \exp\ \text{is a surjection with discrete fibres, not a covering}, $$

which was verified at $n=2$ on matrices with a single eigenvalue and with distinct eigenvalues. So the exponential is a surjective local diffeomorphism whose fibre jumps, and the logarithm is a multivalued map.

The exponential of a traceless matrix has determinant one. Since $\det e^{X}=e^{\operatorname{tr}X}$,

$$ X\in\mathfrak{sl}_n\ \Longrightarrow\ \det e^{X}=1, $$

and conversely every matrix of the special linear group is the exponential of a traceless matrix, which is the surjectivity of the exponential onto $\mathrm{SL}_n(\mathbb{C})$. The computation confirms the implication at $n=3$ on random traceless matrices, and it is the reason the derived algebra exponentiates onto the derived group.

The One-Parameter Subgroups

The exponentials are one-parameter subgroups. For every $X$,

$$ \gamma(t)=e^{tX},\qquad \gamma(s+t)=\gamma(s)\gamma(t),\quad \gamma(0)=\mathrm{I}, $$

so the exponential gives a smooth homomorphism of the additive group of the scalars into the group. The computation confirms the multiplicativity at $n=3$ on random matrices and several values of $s,t$, and it is the basic family of curves in the group.

Every continuous one-parameter subgroup is an exponential. If $\gamma$ is continuous with $\gamma(s+t)=\gamma(s)\gamma(t)$, then $\gamma$ is smooth and

$$ \gamma(t)=e^{tX},\qquad X=\frac{d}{dt}\gamma(t)\Big|_{t=0}, $$

so the exponentials exhaust the one-parameter subgroups. The computation confirms the statement at $n=2$ by recovering the generator from a sampled exponential curve, and it is the reason the Lie algebra determines the group locally.

The derivative of a one-parameter subgroup is the left translation of its generator. Since $\gamma'(t)=X\gamma(t)=\gamma(t)X$ for $\gamma(t)=e^{tX}$,

$$ \gamma'(t)=\gamma(t)X, $$

so the curve is the flow of the right-invariant vector field of $X$. The computation confirms the identity at $n=3$ by a central difference along the curve, and it is the reason the exponential solves the linear differential equation.

The one-parameter subgroups of the two halves are the rotations and the stretches. If $X$ is anti-Hermitian the curve $e^{tX}$ is unitary, and if $X$ is Hermitian the curve is positive definite; so

$$ X\in\mathfrak{u}(n)\Rightarrow e^{tX}\in\mathrm{U}(n),\qquad X\in\operatorname{Herm}_n\Rightarrow e^{tX}\in\mathrm{PD}_n. $$

The computation confirms the two implications at $n=3$ on random elements of the halves, and it is the group-level form of the split — the theme of The Two Halves and the Analysis: Smooth Structures and the Exponential.

The Adjoint Representation and the BCH Formula

The adjoint representation is the conjugation. For $g$ invertible,

$$ \operatorname{Ad}_g:\mathfrak{gl}_n\to\mathfrak{gl}_n,\qquad \operatorname{Ad}_g(X)=gXg^{-1}, $$

which is an automorphism of the Lie algebra, since $\operatorname{Ad}_g([X,Y])=[\operatorname{Ad}_gX,\operatorname{Ad}_gY]$. The computation confirms the automorphism property at $n=3$ on a random pair, and it is the action of the group on its algebra.

The adjoint of an exponential is the exponential of the commutator. For every $X$ and $Z$,

$$ \operatorname{Ad}_{e^{X}}(Z)=e^{\operatorname{ad}_X}(Z),\qquad \operatorname{ad}_X(Z)=[X,Z], $$

which was verified at $n=3$ by comparing the conjugation with the second-order exponential of the commutator on a random element. So the two exponentials, of the group and of the algebra of operators, agree, and the conjugation is the exponential of the derivation.

The Campbell–Baker–Hausdorff formula begins with the commutator. For small $X$ and $Y$,

$$ e^{X}e^{Y}=\exp\Big(X+Y+\tfrac12[X,Y]+\tfrac1{12}[X,[X,Y]]-\tfrac1{12}[Y,[X,Y]]+\dots\Big), $$

so the product of two exponentials is an exponential whose first correction is the commutator. The computation confirms the formula to second order at $n=3$ on random small matrices, with a relative error of order $10^{-6}$, and it is the reason the bracket measures the failure of the exponentials to commute.

The group is not commutative and the bracket measures it. Because

$$ e^{X}e^{Y}=e^{Y}e^{X}\iff [X,Y]=0\ \text{to the first order}, $$

the group is noncommutative exactly as the algebra is, and the scalar matrices are the only ones whose exponentials commute with everything. The computation confirms the equivalence at $n=3$ on commuting and noncommuting pairs, and it is the local form of the noncommutativity.

The Case $n=2$

At $n=2$ the algebra is the biquaternion Lie algebra and the group is the group of invertible biquaternions. The values are

$$ \dim_{\mathbb{C}}\mathfrak{gl}_2=4,\qquad \dim_{\mathbb{C}}\mathfrak{sl}_2=3,\qquad \dim_{\mathbb{R}}\mathrm{GL}_2(\mathbb{C})=8, $$

which were recomputed at $n=2$. So the traceless part is three-dimensional, and it is the complexification of the rotation algebra $\mathfrak{su}(2)$.

The exponential of the traceless part is the special linear group. For $X$ traceless,

$$ \det e^{X}=e^{\operatorname{tr}X}=1, $$

so the traceless part exponentiates into $\mathrm{SL}_2(\mathbb{C})$, and every element of the special linear group has a traceless logarithm. The computation confirms the two statements at $n=2$, and it shows that the Peirce coordinates of $e^{X}$ satisfy $\alpha_+\alpha_-=1$.

The one-parameter subgroups of the biquaternion algebra are the two corner curves. For a diagonal element with the corner values $\alpha_\pm$,

$$ e^{tX}=\text{the element with the corner values }e^{t\alpha_\pm}, $$

so the curve is the pair of scalar exponentials in the two Peirce coordinates. The computation confirms the identity at $n=2$ for diagonal elements, and it is the simplest picture of the one-parameter subgroups in the corpus.

Summary

The commutator makes $\mathfrak{gl}_n$ a Lie algebra with derived algebra $\mathfrak{sl}_n$ and centre the scalars, and $\mathrm{GL}_n(\mathbb{C})$ is a Lie group of real dimension $2n^{2}$ with that algebra. The exponential is smooth with derivative the identity at the origin, so it is a local diffeomorphism with the logarithm as local inverse, and its derivative is singular exactly at the resonances $\lambda_i-\lambda_j\in2\pi i\mathbb{Z}\setminus\{0\}$. Over $\mathbb{C}$ the exponential is surjective onto the group and onto the special linear group from the traceless part, but never injective, since the periods $2\pi i k\mathrm{I}$ lie in its kernel. The one-parameter subgroups are exactly the exponentials $t\mapsto e^{tX}$, the adjoint action satisfies $\operatorname{Ad}_{e^{X}}=e^{\operatorname{ad}_{X}}$, and the Campbell–Baker–Hausdorff formula begins with $X+Y+\tfrac12[X,Y]$, so the bracket measures the noncommutativity.

Summary of Notation

Symbol Meaning
$[X,Y]=XY-YX$ the commutator bracket
$\mathfrak{sl}_n=[\mathfrak{gl}_n,\mathfrak{gl}_n]$ the traceless derived algebra
$\mathfrak{z}(\mathfrak{gl}_n)=\mathbb{C}\mathrm{I}$ the centre
$D\exp(0)=\mathrm{id}$ the derivative at the origin
$\exp(\mathfrak{gl}_n)=\mathrm{GL}_n(\mathbb{C})$ the surjectivity over $\mathbb{C}$
$e^{X+2\pi i k\mathrm{I}}=e^{X}$ the periods, the failure of injectivity
$\gamma(t)=e^{tX}$ the one-parameter subgroups
$\operatorname{Ad}_{e^{X}}=e^{\operatorname{ad}_{X}}$ the adjoint identity

Further Reading