VERY BASIC LIE THEORY ROGER HOWE Department of Mathematics, Yale University, New Haven, CT 06520 Lie theory, the theory of Lie groups, Lie algebras and their applications, is a fundamental part ‘of mathematics. Since World War IT it has been the focus of a burgeoning research effort, and is how seen to touch a tremendous spectrum of mathematical areas, including classical, differential ‘and algebraic geometry, topology, ordinary and partial differential equations, complex analysis (one and several variables), group and ring theory, number theory, and physics, from classical to quantum and relativistic. It is impossible in a short space to convey the full compass of the subject, but we will cite some examples. An early major success of Lie theory, occurring when the subject was still in its infancy, was to provide a systematic understanding of the relationship between Euclidean geometry and the newer geometries (hyperbolic non-Euclidean or Lobachevskian, Riemann’s elliptic geometry, and projective geometry) that had arisen in the 19th century. This led Felix Klein to enunciate his Erlanger Programm [Kl] for the systematic understanding of geometry. The principle of Klein's program was that geometry should be understood as the study of quantities left invariant by the ‘action of a group on a space. Another development in which Klein was involved was the Uniformization Theorem [Be] for Riemann surfaces. This theorem may be understood as saying that every connected two-manifold is a double coset space of the isometry group of one of the 3 (Buclidean, hyperbolic, elliptic) standard 2-dimensional geometries. (See also the recent article [F] in this MONTHLY.) Three-manifolds are much more complex than two manifolds, but the intriguing work of Thurston [Th] has gone a long way toward showing that much of their structure can be understood in a way analogous to the 2-dimensional situation in terms of coset spaces of certain Lie groups. More or less contemporary with the final proof of the Uniformization Theorem was Einstein's [E] invention of the special theory of relativity and its instatement of the Lorentz transformation as a basic feature of the kinematics of space-time. Einstein’s intuitive treatment of relativity was followed shortly by a more sophisticated treatment by Minkowski [Mk] in which Lorentz ‘transformations were shown to constitute a certain Lie group, the isometry group of an indefinite Riemannian metric on R*. Similarly, shortly after Heisenberg [Hg] introduced his famous ‘Commutation Relations in quantum mechanics, which underlie his Uncertainty Principle, Her- mann Weyl (W] showed they could be interpreted as the structure relations for the Lie algebra of a certain two-step nilpotent Lie group. As the group-theoretical underpinnings of physics became better appreciated, some physicists, perhaps most markedly Wigner [We], in essence advocated extending Klein’s Erlanger Programm to physics. Today, indeed, symmetry principles based on Lie theory are a standard tool and a major source of progress in theoretical physics. Quark theory {Dy}, in particular, is primarily a (Lie) group-theoretical construct. These examples could be multiplied many times. The applications of Lie theory are astonishing in their pervasiveness and sometimes in their unexpectedness. The articles of Borel [Bo2] and Dyson [Dy] mention some. The recent article of Proctor {Pr] in this MONTHLY discusses an application to combinatorics. Some points of contact of Lie theory with the undergraduate curriculum are listed in §7. The article of Proctor also illustrates the need to broaden understanding of Lie theory. Proctor did not feel he could assume knowledge of basic Lie theoretic facts. Though hardly an unknown subject, Lie theory is poorly known in comparison to its importance. Especially since it provides Roger Howe received his Ph.D. from the University of California at Berkeley in 1969. His advisor was Calvin C. Moore. He was at the State University of New York at Stony Brook from 1969 to 1974, and since then has been at pa cae His main research interests are in group representation theory and harmonic analysis, both pure applic 6001983) VERY BASIC LIB THEORY 603 ©) If U is a neighborhood of 1, there are other neighborhoods V, W of Vy such thar vw U, where : : (7) VW = (ghigeV,he Ww) given a neighborhood UC G of 1, we must find a neighborhood U” c ox G of 1, ¥ Ig such that for any point (x; 8°) of U’, the product ge” is in U. But by definition of the product topology on GX G, any neighborhood U’ of 1, x 1, contains a product Vx W, where VWOG ae both neighborhoods of 1,,. But the image of V x Wunder the multiplication map is just the set VW defined in £:7- So condition (b) amounts to continuity of multiplication at the point X1ILEGXG * ‘Thus to complete the lemma we need to show that if the multiplication and inverse maps are continuous at the identity, and if the topology on G is homogenteous, then they are continuous everywhere. Let U be a neighborhood of 1. Then since X, is a homeomorphism, gl is a neighborhood of g, and we may write aa) (gu) = wig = p(w“) = p,((Ay-s(eu))") we wu. Thus on gU, the inverse map is a composition of A, the inverse map on U, and p,. Since Ky» and p, are continuous, and A,,-: takes g to 1¢, and the inverse map is continuous at I. we see that the inverse map is continuous at g also. The proof that multiplication is continuous everywhere is analogous and is left as an exercise. @ We return to the question of topologizing Hm( X). Corollary 2 allows us to save work in our efinition of the topology on Hm(X) by only defining neighborhoods of the identity map ly on X, and declaring by fiat all left or right translates of these neighborhoods also to be open sets. Lemma 3 tells us what we must check to know our definition yields a group topology. From now on, we take X to be a locally compact Hausdorff space. Let CC X be compact, and let O 2 C be open. Define (9) U(C,O) = {hE Hm(X):h(C) ¢ O,k-(C) ¢ O}. If (C)}, 1 X be a map. For t € R, define @,: X + X by formula (2.3) (i). Then (,) is a one-parameter group of homeomorphisms if and only if (a) © satisfies identities (2.4) and (b) is continuous, REMARK. According to this lemma, if our goal were simply to define a one parameter group of homeomorphisms in the quickest way, we could short-circuit the whole discussion of $1 and simply define a one-parameter group of homeomorphisms as a map ® satisfying the conditions of the lemma. However, that approach seemed unduly formalistic. Proof. It is a straightforward computation to verify that the identities (2.4) guarantee that for each ¢ the map 9, is in Bi(X) and t + , is a homomorphism. Also it is obvious that the maps @, will be in Hm( X) if and only if @ is continuous in x for each fixed t. Thus the main thrust of the Jemma is that ¢ > , is continuous from R to Hm(X) if and only if ® is jointly continuous in t and x. Let us verify this. Suppose ® is continuous. Let C ¢ X be compact, and O ¢ X be open, with C ¢ O. Choose any x € C. By identity (2.4)()), the point (0, x) € R x X is in O~ 1(0). By continuity of ® a neighborhood of (0, x) is contained in ®~1(0), This means there is a neighborhood N of x in X, and 6 > 0, depending on x, N, and O, such that ®(1, y) € 0 fory © N and|t| < 6. In other words @,(¥) € O for |t| < 8 and y € N. Since C is compact, we can find a finite number of x, € C such that the associated neighborhoods N; cover C. Suppose then that ®(r, y,) € O for y, © N, and |t| <6, Set 6 = miné,. Then we have (1, c) € O for all c € C and |t| <6. In other words, 9, € U(C, O) for |t| < 8. Clearly, by repeating this argument for any finite collection of compact C;'s and open Q;’s containing them, we can show that g, € U({C;},{O,}) for all sufficiently small t. This shows that ¢ > , is continuous at the origin in R. But now we appeal to the following lemma. Lean 7. Let p: G > H be a homomorphism between topological groups. Then @p is continuous if and only if @ is continuous at 1g. The proof of this lemma is left as an exercise to the reader, who will recognize in it the same spirit that informs Lemmas 2, 3, and 5, To finish Lemma 6, we must show that the continuity of ¢ —> g, implies continuity of ® Choose (t,x) ER XX, and set y= (t,x). Let V be a neighborhood of y. Since g, is continuous, we can find a neighborhood W of x, with compact closure W, such that @,(W) & V. Since g, is continuous in 1, we can find e > 0 so that p, € U(g,(W), V) for |s| < e, But then if (',w) © (t— e,1 + &) x W, we have D(1',) = O(t" ~ £, (1, w)) = gy (®(t,)) € Ge<(H(W)) SV. In other words (1 — e, 1 + e) X WC ®-'(V). Since V was an arbitrary neighborhood of », we see ® is continuous at (1, x). Since (t,x) is arbitrary, we see ® is continuous, @ Consider a one-parameter group g, of homeomorphisms of X and the associated map ® defined by formula (2.3), The map ® is a function of two variables, ¢ and x, and the maps 9, are obtained from ® by temporarily fixing ¢ and letting x vary. If on the other hand we fix x and let ¢ vary, we get a map t > (1, x) = g,(x) which defines a continuous curve in X, traced by the moving point ¢,(x). Thus as 1 varies, each point of x moves continuously inside X, and various Points move in a coherent fashion, so that we can form a mental picture of them flowing through X, each point along its individual path. For this reason, a one-parameter group of homeomor- phisms of X is also sometimes called a flow on X.1983] VERY BASIC LIE THEORY. — NI 607 Fig. 1 homeomorphisms of R. For the system of Fig. 1, the map radians In summary, we have seen that a (smooth) one-parameter group of diffeomorphisms of R* yields solutions of a system of differential equations of the form (2.7), and, conversely, a solution (for all time and all x) of system (2.7) yields a one-parameter group. The two constructs, solutions of systems of ordinary differential equations, and one-parameter groups, thus provide two different points of view on the same mathematical phenomenon. In other words, the notion of one-parameter group provides a geometric and global way of looking at the solutions of a system of ordinary differential equations. As such, it suggests ways of attacking and obtaining informa- tion about ordinary differential equations, and it provides a link between systems of ordinary differential equations and more complex geometric objects such as the Lie groups and Lie algebras discussed in the following sections. is just rotation through an angle of t 3. One-Parameter Groups of Linear Transformations In this section, we show how one-parameter groups of linear transformations of a vector space can be described using the exponential map on matrices. Let V be a finite dimensional real vector space. Let End(V) denote the algebra of linear maps from V to itself, and let GL(V) denote the group of invertible linear maps from V to itself. The usual name for GL(V) is the general linear group of V. If V=R", then End(V) = M,(R), the nX n matrices, and GL(V) = GL,,(R), the matrices with nonvanishing determinants. Let || be a norm on V (c.f. (L, [N). In the usual way there is induced an operator norm, also denoted || ||, on End(V). We recall the definition: G.1) A= sup ail 1vEV— )} A End(V). The norm on End¥ makes End(V) into a metric space. Since the determinant is a continuous function on End(V’), we know that GL(V) is an open subset of EndV (see also (3.6) below), so it also is a metric space. DEFINITION. A one-parameter group of linear transformations of V is a continuous homomor- phism 2) M:R = GL(V). Thus M(*) is a collection of linear maps such that (i) M(0) = 1), the identity of V, (i) M(s)M(t) = M(s + t) 5,f€R,Substituting this in our formula for exp A exp B, and noting that all manipulations are valid because the series converge absolutely, we see the proposition follows. i COROLLARY 9. For any A € End V, the map t > exp(tA) is a one-parameter group of linear transformations on V. In particular exp A € GL(V) and (exp(A))~! = exp(—A). Proof. Since for any real numbers s and ¢ the matrices sA and 1A commute with one another, this corollary follows immediately from Proposition 8. @ The main result of this section is the converse of Corollary 9. ‘THEOREM 10. Every one-parameter group M of linear transformations of V has the form (3.10) M(t) = exp(1) for some A € End V. The transformation A is called the infinitesimal generator of the group 1 > exp(1A). The flow illustrated in Figure 1 is in fact given by a one-parameter group with infinitesimal generator Desk i: ol REMARK. Since for v € V we have = fll (ep (exp A)(v) =v + a(v) + E LAO), the infinitesimal generator of the one-parameter group M(1) = exp(1A) can be computed by the formula (Ga) ito) Eis BOW) =e. 0 HOMO). Thus the one-parameter group M(t) is associated by the discussion at the end of §2 to the system of differential equations (3.12) “ =A(v). These equations are of course basic in the theory of linear systems, which is applied in electrical engineering, economics, etc. If we know that M(1)(v) is differentiable, then the existence and uniqueness theorem for differential equations implies Theorem 10, but we do not know a priori that M(1) is differentiable. The burden of the proof of Theorem 3 is to get around this ignorance, thereby establishing that a merely continuous map t > M(t) satisfying the group law (3.2) (ii) is in fact analytic. This is a recurrent theme in Lie theory, and is also expressed in the main theorem (Theorem 17) of this paper. It found its ultimate expression in Hilbert’s Sth Problem: to show that if a topological group is locally (ie., a neighborhood of every point is) homeomorphic to Euclidean space, then the group is in fact an analytic manifold with analytic group law (a Lie group). This problem was resolved positively in the early 1950’s by A. Gleason [G]. See also [Ka], (Mz). We take up now the proof of Theorem 10. It will require some preliminary results. Let @,(A) be the open ball of radius r around A, as defined in formula (3.3). PRoposITION 11. For sufficiently small r > 0, the map exp takes @,(0) bijectively onto an open neighborhood of 1, in GL(V). One has exp(#,(0)) © @,(1y) where s = e" — 1. Proof. Let Dexp, be the differential of exp at A. It is a linear map from End(V) to End(V) defined by " A + tB) — A Dexp,(B) = lim 204 = 1B) — exp From the definition (3.8) of exp, it is easy to compute that Dexp)(B) = B.Neve me se ey Clg1983] VERY BASIC LIE THEORY 611 for all integers m and n. Since the numbers m2-" ae are dense in R, Theorem 10 follows by continuity. 4. Properties of the Exponential Map The map exp is the basic Tink between the linear structure on End V and the multiplicative structure on GL(V). We will describe some salient properties of this link, Choose r with 0 00, we see that nC, > A + B. Since exp nG, = (exp C,)", equation (4.14) fol- lows. @ Recall that the (linear) commutator [A, B] is defined in equation (4.2). Recall also that if g. f are elements of a group, then the group commutator of g and h, written (g: h), is the expression (gh) = ghg ‘a’. Proposition 15 (Commutator formula). For A, B © End V, one has (415) exp[ A, B] = lim (exp(/n)exp(B/n)exp(—A /n)exp(—B/n))" = lim ((exp(A/n) :exp(B/n))” Proof. As in Proposition 14, for large n we have exp(A/n)exp( B/n) = exp C, = ovo((4 + BY/nty taal + 5) where uss < os flAl He Similarly exp(—A/n)exp(-B/n) = exo (= (4 + B)/n +(5) 421 “* si) = exp G,1983] VERY BASIC LIE THEORY 615 xp(—tA)) = B of (B) = tim (P14) B(e a t c $ (exp (4) B(exp(~1A))],.9 = (A(exp 1A) B(exp ~ 1A) + (exp 4) B(—A)(exp = 1A))|,-9 = AB- BA =[A,B). Here we have used the fact that d Spex) = A exp(1A) This formula may be verified by direct calculation from the definition of exp(t4). Hence if we define ad 4: End V + End V by adA(B) =[A, B], we have the following formula. PROPOSITION 17. For A € End V (4.17) Ad(exp A) = exp(ad A). 5. The Lie Algebra of a Matrix Group By a matrix group we mean a closed subgroup of GL(V) for some vector space V. This section shows a matrix group is a Lie group. What that means is expressed in Theorem 17. Most, though not all, Lie groups can be realized as matrix groups. This article discusses only matrix groups. Exampes. (i) GL(V) itself. (i) SL,(R), the special linear group, of x n matrices of determinant 1. (Gi) O,,,, the “pseudo-orthogonal groups,” consisting of all matrices in GL,.,,(R) that preserve the indefinite inner product P pa (%,2)5.4= Dx > DL mx, x xe RP, ist impet (iv) SP,,(R), the real symplectic group, consisting of all matrices in SL.,(R) that preserve the skew-symmetric bilinear form x0) = Dixon ~ XX pene x" ER ded (v) The group P(U) of transformations that preserve a subspace U of V. For instance, if V=R", and U,, =R” = {(X1, X25.++5% 15040)... 50)}, Where m 0 exp(p2(4n)) = eP( P1(4n)) “exp B, a PNA 9 0 pti P by our assumption on B,, nO; Pal) 0 a, js also in G, and is (A wil have cluster points, and these must be in q by Lemma 1g" sequence ip3(4.)Il_ P2(4*" Ty cluster points must be in ¥. This contradicts the fact that asl “git bq, so statement (b) of Theorem 17 follows. m sen ExAu i ie algebras of some of the gro sil describe below the Lie algebras ; ae ie ee eat pees tie verification that the indicated Lie algebras are indeg, algebras of the stated groups is left as an exercise. i V). i ie algebra of GL(V) is of course End( i 5 as ne Sere of SL,(R) is the space of sl, (R) of n X n matrices of trace zero Gi) Let Abe a bilinear form on V. The isometry group of Bis the group of invertible at A such that re M0 the Y was dat the d the Lig B(Au, Av) =B(u,v) forall = uyveV, The Lie algebra of this group is the space of operators B such that B(Bu, v) + B(u, Bo) = 0. In particular the Lie algebra 0,(V) of the orthogonal group 0,,(R) of isometries of the standard inner product on R” is the space of skew-symmetric matrices, (iy) The Lie algebra of the subgroup of GL(V) of maps commuting with given operators {T) is the subalgebra of End V commuting with the T,. (¥) The Lie algebra of the group P({¥;}) of invertible transformations which preserve each of the subspaces V, of V is the subalgebra of all transformations which preserve the V,, In Particular, the Lie algebra of the group of invertible upper triangular matrices is the vector space of all upper triangular matrices. (vi) The Lie algebra of G, O G;, for matrix groups G,, is a, 0 ¢>. (vii) A matrix group G and its identity component G° have the same Lie algebra. Alter its existence, the second most important feature of g is that it is natural (in the sense of category theory). This is the content of our next theorem. Let 4, 5, be real Lie algebras. A homomorphism from g to 6 is a linear map Lig>5 satisfying (5.3) L([x,y])=[Lx, Ly] x,y ea. Let V, U be real vector spaces. THEOREM 19. Let GC GL(V) be a matrix group with Lie algebra g. Let ¢:G + GLU) bea continuous homomorphism. Then there is a homomorphism of Lie algebras (5.4) do:g > EndU such that i) exp(d$(A)) = $ (exp A). Proof. IfA € g, then exp 14 is a one isa one-parame” ter. of ‘Parameter subgroup of G, so $(exp(tA)) is 4 B) {0 same ee eee . Hence by Theorem 10 we may write $(exp(tA)) = ¢*”" d$(A) = B,1983] VERY BASIC LIE THEORY 621 describes conjugacy classes in U,, the n x n unitary group. We cannot explain this i i ; . thi of these results in detail, but encourage the reader to explore it by further reading Aso, SL(R) or SL,(C) are examples of an extremely important class of Lie groups called semisimple groups, and several well-known results in linear algebra are special cases for SL, (R) or sL.(0) of waa oe semisimple groups. (Since GL, and SL, are so similar, we state the results for GL,,.) The polar decomposition or singular value decompocith : A € GL,(R) may be written in the form eee ea A= 0S =0,D0,, where O, O,, and O; are orthogonal matrices, S is symmetric, and D is diagonal with positive entries. This isthe specialization to GL.,(R) of what is known as the Cartan decomposition [He] in the context of semisimple Lie groups. Also, the Gram-Schmidt orthonormalization procedure [St] says, in group-theoretical terms, that any A € GL,,(R) may be written in the form A= OB=oDU, where 0 is orthogonal, B is upper triangular, D is diagonal with positive entries, and U is upper triangular with diagonal entries all equal to 1. For general semisimple groups, this is known as the Iwasawa decomposition [He]. Various basic features in the elimination theory, including the “LU factorization” [St] of a generic matrix into the product of an upper triangular and a lower triangular matrix, and the “reduced row-echelon form” [DN] are aspects of a different kind of decomposition of semisimple groups, known as the Bruhat decomposition (Bo). 2. The cross product on R? defines a Lie algebra structure on R°. This is in fact isomorphic to 05, the Lie algebra of O,, the 3 x 3 skew symmetric matrices. The isomorphism is accomplished by 0 -x -y Gy2) |x 0 <2 ceed) The generalization of this correspondence to higher dimensions leads to the theory of spinors and Clifford algebras [J2]. 3. The fact that second mixed partial derivatives are equal is a reflection of the fact that R” is an abelian Lie group. 4, The theory of Fourier series and Fourier transform is best understood group-theoretically. See [Gr] for a discussion. 5. It is fairly routine in quantum mechanics courses, in conjunction with the Schrodinger equation for the hydrogen atom and angular momentum, to introduce, “raising and lowering operators” [Me]. The operators belong to the complexification of 03, which is isomorphic to sl,(C). The commutation relations of the Lie algebra figure importantly in the computations. The harmonic oscillator is also susceptible to a Lie-theoretic treatment. The Canonical Commutation Relations themselves are the laws for a bracket relation on a Lie algebra, known as the Heisenberg Lie algebra [Ca], [Ho]. The relations of this algebra with quantum mechanics, and physics generally, is deep and extensive. 6. Pethaps the part of standard undergraduate mathematics that is pedagogically most ‘compatible with Lie theory is differential equations. We have already discussed in §2 how the notion of one-parameter group is a geometrization of the solution of a system of differential equations. And in §3 we noted that one-parameter groups of linear transformations wire associated with the very important class of linear, constant coefficient systems. Indeed, fphovenial map and linear algebra techniques are often expisily wed in weating these Seem