$\proof$
As $f$ is a homomorphism we have for all $x,a\in G$: $f(x)=f(a^{-1})f(ax)$, i.e. $f=L_{f(a^{-1})}\circ f\circ L_{a}$; hence $f_{*}=L_{f(a^{-1})*}\circ f_{*}\circ L_{a*}$ and thus we conclude that $f_{*e}=L_{f(a^{-1})*}\circ f_{*a}\circ L_{a*}$. Since both $L_{a*}$ and $L_{f(a^{-1})*}$ are isomorphisms the linear mappings $f_{*a}$ and $f_{*e}$ have the same rank. The conclusion follows from the Rank Theorem of differential geometry.
$\eofproof$
Suppose $X$ is a left invariant vector field and $X_e\in T_eG$, then by definition of left invariance: $X_g=L_{g*}(X_e)$. Hence $X$ is uniquely defined by providing $X_e$, i.e. ${\cal G}$ may be identified with the tangent space $T_eG$ of $G$ at $e$. Since $L_{g*}([X,Y])=[L_{g*}(X),L_{g*}(Y)]$, the commutator $[X,Y]$ of two left invariant vector fields is again left invariant. Thus $({\cal G},+,[.,.])$ is a Lie-algebra, i.e. $[.,.]$ is bi-linear, skew symmetric and satisfies Jacobi`s-identity:
$$
[X,[Y,Z]]+[Y,[Z,X]]+[Z,[X,Y]]=0~.
$$
$\proof$
Of course we must have $Y_{e^\prime}=F_*(X_e)$ and thus the only left invariant vector field on $H$ must be $Y_h\colon=L_{h*}(F_*(X_e))$. Thus we are left to prove that $X$ and $Y$ are $F$-related. Since $F$ is a homomorphism we have $F\circ L_g=L_{F(g)}\circ F$ and therefore
\begin{eqnarray*}
F_*(X_g)
&=&F_*(L_{g*}(X_e))
=(F\circ L_g)_*(X_e)\\
&=&(L_{F(g)}\circ F)_*(X_e)
=L_{F(g)*}\circ F_*(X_e)
=Y_{F(g)}~.
\end{eqnarray*}
$\eofproof$
For $G=\Gl(n,\R)$ we have ${\cal G}=\Ma(n,\R)$ is the set of all $n\times n$ matrices; moreover $[X,Y]=XY-YX$. As $L_AB\colon=AB$ we get: $L_{A*}(B)U=DL_A(B)=AU$. For each $X\in\Ma(n,\R)$ the only left invariant vector field $X_A$ satisfying $X_1=X$ is given by $X_A\colon=DL_A(1)X=AX$. The flow of this vector field is $\theta(t,A)=A\,e^{tX}$.
$\proof$
Since $X$ is left invariant we have for all $t\in\R$ and all $g\in G$: $L_g\circ\theta_t=\theta_t\circ L_g$, i.e. $g\theta(t,x)=\theta(t,gx)$. Thus it suffices to verify that $c(t)\colon=\theta(t,e)$ is defined on all of $\R$. By uniqueness and existence of the local flow $c(t)$ is defined for $|t| < \e$. But $c$ is a homomorphism on $(-\e,\e)$ and thus it admitts a unique extension on all of $\R$: just put for any $k\in\Z$: $c(kt)\colon=c(t)^k$. Obviously $X_e$ is the generator of this extension and as for any flow: $\theta_s\circ\theta_t=\theta_{s+t}$ we get:
$$
\theta_e(s+t)
=\theta(s+t,e)
=\theta_s(\theta_t(e))
=\theta(s,\theta_t(e))
=\theta_t(e)\theta_s(e)~.
$$
Conversely assume $c:\R\rar G$ is a one parameter sub-group of $G$ and $X$ a left invariant vector field satisfying $X_e=c^\prime(0)$. Put $\theta(t,g)\colon=gc(t)$, then for all $h\in G$: $L_h\circ\theta_t=\theta_t\circ L_h$; denoting by $Y$ the generator of $\theta$, we see that $Y$ is left invariant and as $Y_e=X_e$ we infer that: $X=Y$.
$\eofproof$
$\proof$
Let $X^1,\ldots,X^n$ be left invariant vector fields $G$ such that ${\cal H}=\lhull{X^1,\ldots,X^n}$. Let $\D$ be the corresponding distribution; by assumption $\D$ is an involutional distribution and thus by koms9 there is a maximal integral manifold $H$ containing the neutral element $e$. Now for all $g\in G$ the left translation $L_g$ is a diffeomorphism on $G$ and thus $L_g(H)$ is a maximal integral manifold through $g$. On the other hand we clearly have for all $h\in H$: $L_{h^{-1}}(H)\cap H\neq\emptyset$ i.e. $L_{h^{-1}}(H)=H$, which shows that $H$ is a sub-group of $G$. As the mappings $m:H\times H\rar H$, $m(x,y)\colon=xy$ and $i:H\rar H$, $i(x)=x^{-1}$ are smooth - being restrictions of smooth mappings, $H$ is a Lie-group.
$\eofproof$
If two left invariant vector fields commute their flows commute as well, i.e. $[X,Y]=0$ implies $\exp(X+Y)=\exp(X)\exp(Y)$. In general we have the BCH-formula:
\begin{equation}\label{lgreq1}\tag{LGR1}
\exp(X)\exp(Y)
=\exp\Big(X+Y+\tfrac12[X,Y]+\tfrac1{12}[X,[X,Y]]+\tfrac1{12}[Y,[Y,X]]+\cdots\Big)
\end{equation}
for all $X,Y\in T_eG$ in a sufficiently small neighborhood of $0$. Exploiting the fact that $\exp:T_eG\rar G$ is a local diffeomorphism we get by results of chapter:
$\proof$
Let $B$ be an open ball in $T_eG$ about $0$ such that $\exp:B\rar U$ is a homeomorphism. By corollary we have $\bigcup_{n=1}^\infty U^n=G$ and as $G$ is commutative we also have $\exp(X^1)\exp(X^2)\cdots\exp(X^l)=\exp(X^1+\cdots+X^l)$. Hence $\exp:T_eG\rar G$ is a continuous, surjective homomorphism. By corollary we conclude that $G$ is isomorphic to $T_eG/\ker\exp$. Now $\exp$ is a local diffeomorphism and thus $\ker\exp$ must be a discrete subgroup of $T_eG\simeq\R^n$. By proposition $\ker\exp$ is isomorphic to $\Z^k$, which implies that $T_eG/\ker\exp$ is isomorphic to $\R^n/\Z^k\simeq\TT^k\times\R^{n-k}$.
$\eofproof$
Differential forms and associated measures
1. Given two separable metric spaces $M,Y$, an open covering $U_j$ of $M$ and homeomorphisms $\vp_j:U_j\rar Y$ we define $\Phi_{jk}:\vp_j(U_j\cap U_k)\rar\vp_k(U_j\cap U_k)$ by $\Phi_{jk}\colon=\vp_k\circ\vp_j^{-1}$. Now suppose we have for each $j$ a measure (or finite signed measure) $\l_j$ on $Y$, such that the image measure of the restriction $\l_j|\vp_j(U_j\cap U_k)$ under $\Phi_{jk}$ coincides with the restriction $\l_k|\vp_k(U_j\cap U_k)$ - in this case we simply write: $\l_k=\Phi_{jk}\#\l_j$, i.e.
$$
\l_k(A)=\l_j(\Phi_{jk}\in A)
\quad\mbox{or}\quad
\l_k(\Phi_{jk}(A))=\l_j(A)~.
$$
Then there is exactly one measure $\mu$ on $M$ such that the image measure of $\mu|U_j$ under $\vp_j$ equals $\l_j$: Indeed denoting by $\mu_j$ the measure $\mu_j(U)\colon=\l_j(\vp_j(U))$ on $U_j$, we get for any open subset of $U_j\cap U_k$:
$$
\mu_j(U)
=\l_j(\vp_j(U))
=\l_k(\Phi_{jk}(\vp_j(U))
=\l_k(\vp_k(U))
=\mu_k(U)~.
$$
Hence $\mu_j$ and $\mu_k$ coincide on $U_j\cap U_k$. For any measurable subset $A$ of $M$ we finally put
$$
A_k\colon=(A\sm\bigcup_{j < k}U_j)\cap U_k
\quad\mbox{and}\quad
\mu(A)\colon=\sum\mu_k(A_k)~.
$$
2. Assume $Y=\R^n$, let $M$ be an $n$-dimensional, orientable manifold and let $(U_j,\vp_j)$ be an orientable atlas such that all transition mappings $\Phi_{jk}$ are diffeomorphisms satisfying: $\det D\Phi_{jk} > 0$. Moreover we will assume that $\l_j$ is a signed measure absolutely continuous with respect to Lebesgue measure $\l$ on $\R^n$: the density will be denoted $\r_j$. By the substitution rule we have $\l_k=\Phi_{jk}\#\l_j$ if and only if:\begin{equation}\label{lgreq2}\tag{LGR2}
\r_j=\r_k\circ\Phi_{jk}\det D\Phi_{jk}
\end{equation}
where $D\Phi_{jk}$ denotes Jacobi`s matrix of $\Phi_{jk}$.
3. Suppose finally that $\o$ is an $n$-form on the manifold $M$ with compact support, then $\o_j\colon=\vp_j^{-1*}\o$ is an $n$-form on $\R^n$. Hence there is a unique smooth function $\r_j$ such that $\o_j=\r_jv$, where $v$ denotes the volume form $dx_1\wedge\ldots\wedge dx_n$ on $\R^n$ with coordinates $x_1,\ldots,x_n$. We claim that there is exactly one signed measure $\mu_{\o}$ on $M$ such that for all measurable $A\sbe U_j$:
$$
\mu_{\o}(A)=\int_{\vp_j(A)}\r_j(x)\,dx~.
$$
This is because $\o_j=\Phi_{jk}^*\o_k$ and therefore by the formula for the pull back of $n$-forms on $\R^n$:
$$
\r_j=\r_k\circ\Phi_{jk}\det D\Phi_{jk}~.
$$
Hence the functions $\r_j$ satisfy the requirements in 2. and thus $\mu_\o$ is uniquely determined. Obviously, for any $g\in C_c^\infty(M)$ we have: $d\mu_{g\o}=g\,d\mu_{\o}$. The signed measure $\mu_\o$ is said to be the measure associated to the $n$-form $\o$. For measures the construction works on any manifold, oriented or not, by replacing \eqref{lgreq2} with $\r_j=\r_k\circ\Phi_{jk}|\det D\Phi_{jk}|$.
Let $F:N\rar M$ be such that the pull back $F^*:{\cal D}(M)\rar{\cal D}(N)$, $\vp\mapsto\vp\circ F$ is continuous, then its dual $F_\sharp:{\cal D}^\prime(N)\rar{\cal D}^\prime(M)$ is a continuous linear map. If $F$ is a diffeomorphism and $\mu$ has a density with respect to Lebesgue measure, then $F_\sharp(T_\mu)=T_{\mu_{F^{-1}}}$.
As $\mu$ is the image measure of $\mu_{F^*\o}$ under $F$, $\mu_{F^*\o}$ is the image measure of $\mu$ under $F^{-1}$. Hence
$$
\la\vp,T_{\mu_{F^{-1}}}\ra
=\int_N\vp\,F^{-1*}\o
=\int_N\vp\circ F\circ F^{-1}\,F^{-1*}\o
=\int\vp\circ F\,\o
=\int F^*(\vp)\,d\mu
=\la F^*(\vp),T_\mu\ra
=\la\vp, F_\sharp(T_\mu)\ra~.
$$
Suppose we are given two $n$-forms $v$ and $\vol{}$ on $M$. Then for some smooth function $\r:M\rar\R$: $\vol{}=\r\,v$. Hence the measure $\mu_{\vol{}}$ has density $\r$ with respect to $\mu_v$. To determine $\r$ at a point $m\in M$ we choose a basis $X_m^1,\ldots,X_m^n$ for $T_mM$ such that $v(X^1,\ldots,X^n)=1$, then it follows that $\r(m)=\O(X_m^1,\ldots,X_m^n)$. Let`s have a look at some particular cases:
1. $M$ is an open subset of $\R^n$ and $\la.,.\ra$ any Pseudo-Riemannian metric on $M$ with corresponding volume form $\vol{}$ - i.e. $\vol{}(X^1,\ldots,X^n)=1$ for any orthonormal basis $X^1,\ldots,X^n$. The measure corresponding to the volume form $v\colon=dx_1\wedge\ldots\wedge dx_n$ is the Lebesgue measure $\l$, i.e. $\mu_v=\l$. Hence the measure $\mu\colon=\mu_{\vol{}}$ corresponding to the volume form $\vol{}$ has density $\r(x)\colon=\vol x(\pa_1,\ldots,\pa_n)$ with respect to $\l$. We would like to calculate this density by means of the inner product $\la.,.\ra$: If $E_x^j$ denotes an orthonormal basis for $T_xM$ such that $\e_j=\la E^j,E^j\ra$ and $\vol{}(E_x^1,\ldots,E_x^n)=+1$, then
$$
\pa_k=\sum_l\e_l\la\pa_k,E^l\ra E^l=\colon\sum_l a_{lk}E^l
$$
and therefore we get: $\r(x)=\det A > 0$ where $A=(a_{jk})$. However we need to determine an orthonormal basis! To work around this problem we start with the Gramian $(g_{jk})$, where $g_{jk}\colon=\la\pa_j,\pa_k\ra$. Next we express the coefficients $g_{jk}$ in terms of the entries $a_{jk}$:
$$
g_{jk}\colon=\la\pa_j,\pa_k\ra=\sum_l\e_la_{jl}a_{kl}~.
$$
Finally we notice that the numbers $\sum_l a_{jl}a_{kl}$ are just the entries of the matrix $AA^t$ and $\sum_l a_{jl}a_{kl}\e_l$ are the entries of the matrix $ADA^t$, where $D=diag\{\e_1,\ldots,\e_n\}$. Thus denoting by $\nu$ the index of $\la.,.\ra$ and by $G$ the number $G\colon=\det(g_{jk})$, we conclude: $\det(g_{jk})=(-1)^\nu\det(A)^2$, i.e. $\r=\sqrt{|G|}$ and therefore
\begin{equation}\label{lgreq3}\tag{LGR3}
\forall f\in C_c^\infty(M):\qquad
\int_M f\,\vol{}
=\int_M f\sqrt{(-1)^\nu\det(\la\pa_j,\pa_k\ra)}\,d\l~ .
\end{equation}
2. If $M$ is any orientable Pseudo-Riemannian manifold, $(U,\vp)$ some chart and $\supp(f)\sbe U$, then
$$
\int f\,\vol{}=\int_{\R^n}f\circ\vp^{-1}\sqrt{(-1)^\nu\det(g(\pa_j,\pa_k))}\,d\l
$$
where $g\colon=\vp^{-1*}$ denotes the pull back of metric tensor $\la.,.\ra$ on $U$ by $\vp^{-1}:\R^n\rar U$.
3. Suppose $\vol{}$ denotes the volume form on an orientable Pseudo-Riemannian manifold $M$ and $F:M\rar M$ denotes an orientation preserving diffeomorphism. Then $F^*\vol{}$ is a further volume form on $M$ and hence $F^*\vol{}=f\,\vol{}$, where $f(m)=\det(T_mF)=F^*\vol m(E^1,\ldots,E^n)$ for any positively oriented orthonormal reper $E^j$. Putting again $g_{jk}\colon=\la T_mF(E^j),T_mF(E^k)\ra$ and $G\colon=\det(g_{jk})$, we get as above: $f=\sqrt{|G|}$, i.e.
\begin{equation}\label{lgreq4}\tag{LGR4}
F^*\vol{}=\sqrt{(-1)^\nu G}\,\vol{}~ .
\end{equation}
The adjoint representation
Given a smooth automorphism $u:G\rar G$ we can define for any left invariant vector field $X$ another left invariant vector field $u_*X$ by $u_*X_{u(g)}\colon=u_*(X_g)$ - or $u_*X_e=u_*(X_e)$ This way we get an automorphism $u_*:\GG\rar\GG$, $X\mapsto u_*X$. Choosing in particular the inner automorphism $u(x)\colon=i_g(x)\colon=gxg^{-1}$, we get an automorphism $X_e\mapsto i_{g*}(X_e)$; the mapping $g\mapsto i_{g*}\in\Gl(\GG)$ is called the adjoint representation and denoted by $\Ad$, i.e.
\begin{equation}\label{lgreq5}\tag{LGR5}
\Ad(g)(X_e)\colon=i_{g*}(X_e)~.
\end{equation}
Since $i_{gh}=i_g\circ i_h$ this is indeed a representation of the Lie-Gruppe $G$, i.e. $\Ad(gh)=\Ad(g)\Ad(h)$ and $\Ad(e)=id$. In case of a matrix Lie-group $G$ we re-obtain $\Ad(g)X=gXg^{-1}$, cf. section.
Now let $X,Y$ be left invariant vector fields on the Lie-group $G$ and denote by $\theta_t(g)=g\theta_t(e)=R_{\theta_t(e)}(g)$ its flow through $g$. By vs3 we have (remember $\theta_{t*}X_g\colon=\theta_{t*}(X_{\theta(-t,g)})\in T_gG$):
$$
[X,Y]_e=\ttdl t0\theta_{-t*}Y_e~.
$$
Since $Y_g=L_{g*}(Y_e)$ we conclude that:
$$
\theta_{-t*}Y_e
=\theta_{-t*}(Y_{\theta_t(e)})
=\theta_{-t*}(L_{\theta_t(e)*}(Y_e))
=T_e\left(\theta_{-t}\circ L_{\theta_t(e)}\right)(Y_e)
$$
Next notice that $\theta_t=R_{\theta_t(e)}$ and therefore:
$$
\theta_{-t*}Y_e
=T_e\left(R_{\theta_t(e)^{-1}}\circ L_{\theta_t(e)}\right)(Y_e)
=\Ad(\theta_t(e))(Y_e)~.
$$
Putting $\ad\colon=T_e\Ad$ we finally get:
\begin{equation}\label{dmleq1a}\tag{LGR3}
[X,Y]_e=\ttdl t0\Ad(\theta_t)Y_e=\ad(X_e)Y_e~.
\end{equation}
$\proof$
For simplicity let us assume $p=1$. Consider
$$
A_g(X_g)\colon=u(L_{g^{-1}*}(X_g))
\quad\mbox{and}\quad
B_g(X)\colon=u(R_{g^{-1}*}(X))~.
$$
Then we have
$$
L_h^*A_g(X_g)
=A_{hg}(L_{h*}(X_g))
=u(L_{g^{-1}*}(X_g))
=A_g(X_g)~.
$$
Hence $A$ is left invariant. Similarly $B$ is right invariant and both coincide if and only if $A$ is also right-invariant, i.e. for all $g\in G$:
$$
A=R_{g^{-1}}^*A=R_{g^{-1}}^*L_g^*\Phi=i_g^*A,
$$
in particular: $u=A_e=\Ad(g)^*A_e=\Ad(g)^*u$. Conversely assume that this condition holds, then
\begin{eqnarray*}
A_g(X_g)
&=&u(L_{g^{-1}*}(X_g))
=\Ad(g^{-1})^*u(L_{g^{-1}*}(X_g))\\
&=&u(i_{g^{-1}*}L_{g^{-1}*}(X_g))
=u(R_{g^{-1}*}(X_g))=B_g(X_g)~.
\end{eqnarray*}
$\eofproof$
By results of section every lie-group admits an essentially unique Haar measure. The following results prove that any Haar measure on a Lie-group comes from a volume form!
$\proof$
We start with any euclidean product $g$ and any volume form $v$ on $T_eG$. By lemma there is a unique Riemannian metric $\la.,.\ra$ and a unique volume form $\vol{}$ on $G$, whose restriction to $T_eG$ coincide with $g$ and $v$, respectively.
$\eofproof$
On a compact metric group corollary assures the existence of a bi-invariant metric. In case of a compact Lie-group the subsequent result shows that this can be chosen to be a bi-invariant Riemannian metric.
$\proof$
1. Let $v$ any non-trivial $n$-form on $T_eG$. As the dimension of all $n$-forms on $T_eG$ is one, we must have for all $x\in G$: $\Ad(x)^*v=f(x)v$. Moreover $f:G\rar\R\sm\{0\}$ is a homomorphism because $\Ad(g)$ is. This shows that $f(G)$ is a compact and connected sub-group of the multiplicative group $\R\sm\{0\}$, i.e. $f(G)=\{1\}$. By lemma there is a unique bi-invariant volume form $\vol{}$ on $G$ such that $\vol{e}=v$.
2. Choose any left-invariant Riemannian metric $(.,.)$ and a bi-invariant volume form $\vol{}$ on $G$. For any pair of left-invariant vector fields $X,Y$ we put
$$
\forall x\in G:\quad
f(x)\colon=(\Ad(x)X_e,\Ad(x)Y_e)
$$
and on $T_eG$ we define a euclidean product by
$$
\la X_e,Y_e\ra\colon=\int_Gf\,\vol{}~.
$$
We are left to check that the unique metric tensor extending $\la.,.\ra$ is bi-invariant. By lemma this amounts to proving that $\la\Ad(g)X_e,\Ad(g)Y_e\ra=\la X_e,Y_e\ra$: By definition we have
\begin{eqnarray*}
\la\Ad(g)X_e,\Ad(g)Y_e\ra
&=&\int_G(\Ad(x)\Ad(g)X_e,\Ad(x)\Ad(g)Y_e)\,\vol{}\\
&=&\int_G(\Ad(xg)X_e,\Ad(xg)Y_e)\,\vol{}
=\int_G f\circ R_g\,\vol{}~.
\end{eqnarray*}
and as $\vol{}$ is right-invariant we are done.
$\eofproof$