Wisconsin Bourbaki Seminar

Arun Ram
Department of Mathematics and Statistics
University of Melbourne
Parkville, VIC 3010 Australia
aram@unimelb.edu.au

Last update: 24 March 2014

Notes and References

This is an excerpt from notes of the Wisconsin Bourbaki Seminar, Fall 1993. The notes were written by Oliver Eng, Susan Hollingsworth, Mark Logan, Arun Ram and Louis Solomon.

§2

  1. Let K be a commutative ring with unit, let E be a K-module, and let E* be the dual of E. Let ϕ denote the canonical homomorphism from EE* to End(E).
    1. A pseudo-reflection of E is a non-identity element of End(E) of the form sx,y*=1-ϕ (xy*), where xE and y*E*. A pseudo-reflection s is called a reflection if one may choose x,y* such that x,y*=2; show that then one has that s2=1 and that s(x)=-x.
    2. Let xE, y*E* such that x,y*=1 and let s be the reflection corresponding to the pair (2x,y*). Show that E is the direct sum of the submodule Kx generated by x and the orthogonal H of y*. Show that Kx is free with base x, and that s is equal to 1 on H and to -1 on Kx.

    Solution.
    1. The map ϕ is given by ϕ: EE* End(E) xy* x·,y*. Since , is bilinear sx,y*2 = (1-ϕ(xy*)) (1-ϕ(xy*)) = 1-2ϕ(xy*)+ ϕ(xy*) (ϕ(xy*)) = 1-2ϕ(xy*)+x x·,y*,y* = 1-2ϕ(xy*)+ x·,y* x,y* = 1-2ϕ(xy*)+ ϕ(xy) x,y* = 1-ϕ(xy*) (x,y*-2) or any pseudo-reflection sx,y*. Thus, with the assumption that x,y*=2, we have that sx,y*2=1.
      From the definition of pseudo-reflection we have that sx,y*(x)= [1-ϕ(xy*)] (x)=x-xx,y*. Thus, if x,y*=2, we have that sx,y*(x)=-x.
    2. Let H={hE|h,y*=0} and Kx={kx|kK}.
      We show that if x,y*=1 then
      1. Kx+H=E.
      2. KxH=0.
      3. Kx is free with basis x.
      4. s2x,y*(kx)=-kx for any kK.
      5. s2x,y*(h)=h for any hH.
      1. Let eE. Then e=e,y*x+(e-e,y*x). Since e-e,y*x,y* = e,y*- e,y* x,y* = e,y* (1-x,y*) = 0, we have that e-e,y*xH. It follows that eKx+H.
      2. Let eKxH. Write e=kx for some kK. Then 0=e,y*= kx,y*=k 1=k. So k=0 and e=kx=0.
      3. To show that Kx is free with basis x we must show that every element can be written uniquely in the form kx for some kK. To see this, suppose that kx=kx for some k,kK. Then 0=0,y*= (k-k)x,y*= (k-k)x,y* =k-k. So k=k.
      4. Let kxKx. Then, since x,y*=1, s2x,y*(kx)= kx-kx,y* 2x=kx-k2x=-kx.
      5. Let hH. Then s2x,y*(h) =h-h,y* 2x=h-0=h.
  2. With the notations as in exercise 1, show that det(sx,y*)=1-x,y* if E is a free K-module of finite type.

    Solution. Let e1,e2,,en be a basis of E. Then, for any uEnd(E), u(e1) u(e2) u(en)=det (u)(e1e2en). (*) We have that sx,y* (e1e2en) = (e1-e1,y*x) (e2-e2,y*x) (en-en,y*x) = e1en-i e1ei-1 ei,y*x ei+1en+0. Substituting x=c1e1+c2e2++cnen gives that sx,y* (e1e2en) = e1en +iei,y* e1ei-1 (c1e1+c2e2++cnen) ei+1en = e1en-i ei,y* e1ei-1 cieiei+1 en = e1en-i ciei,y* e1ei-1 eiei+1 en = e1en- x,y* e1en = (1-x,y*) e1e2en. Thus det(sx,y*)= 1-x,y*.
  3. Let V be a complex Hilbert space with basis e1,,e. For 1i, let si be a unitary pseudo-reflection such that si(ei)=ciei where ci1. An element of V is invariant under s if and only if it is orthogonal to ei. Let W be the subgroup of GL(V) generated by the si.
    1. Let i be an integer 1. Show that every element of iV invariant under W is zero.
    2. Suppose that W is finite. Show that for all endomorphisms A of V one has: wWdet (A-w)=Card (W)det(A), wWdet (1-Aw)=Card (W). Deduce that, for all AEnd(V) there exists wW such that Aw does not have any nonzero fixed point.
    3. Let Γ be the graph with vertices in the set [1,] and edges (i,j) determined by the pairs ei,ej such that ei and ej are not orthogonal. Show that V is a simple W-module if and only if Γ is connnected and nonempty.
    4. Suppose that V is a simple W-module. Show that the W-modules iV (0i) are simple. Show that these modules are pairwise nonisomorphic.

    Solution. Let us first see that an element of V is invariant under si if and only if it is orthogonal to ei. Suppose vV is such that si(v)=v. Then, since si is unitary, v,ei= si(v),si(ei)= v,ciei= civ,ei. Since ci1 it follows that v,ei=0. So v is orthogonal to ei. To prove the converse, suppose that v is orthogonal to ei. Then, since si is a pseudoreflection, (1-si)v= c(1-si)ei= c(1-ci)ei for some c. Thus (1-si)v,ei =c(1-ci) ei,ei. On the other hand, (1-si)v,ei = v,ei- siv,ei = 0-ci-1 siv,siei = ci-1 v,ei = 0. Since ci1 and ei,ei0 it follows that c=0. So (1-si)v=0 and thus siv=v.
    1. The proof is by induction on =dimV. When dimV=1 then e1 is a basis of V. Suppose that wce1=ce1, for all wW. Then s1ce1=ce1 c1ce1c (1-c1)e1=0. Since c10 we have that c=0. Now suppose that dimV=. Let V be the subspace spanned by {e1,e2,,e-1} and let e(V). Let W be the group generated by s1,s2,,s-1. If viV then v=a+be,where aiVandb iV. Suppose that viV and that wv=v for all wW. If wW then wv = w(a+be) = wa+wbwe = wa+wbesince e(V) = a+besincewv=v. So wa=a and wb=b for all wW. By induction we have that a=0 and b=0. So v=a+be=0.
    2. We use the determinant relation given in (*). det(A-w)= (A-w) (e1e)= I[1,]± ( Aei1 Aeikwej1 wejm ) , where the sum is over all subsets I={i1,i2,,ik} of [1,]={1,2,,} and Ic={j1,j2,,jm} is the complement of I in [1,]. Then wWdet(A-w) = wWI[1,] ± ( Aei1 Aeikwej1 wejm ) = I[1,]± ( Aei1 Aeik ( wW wej1wejm ) ) . Then, by part a), provided m1, wWwej1wejm=0 since the left hand side is an invariant element of mV. So wW det(A-w) = wWAe1 Ae+ I[1,]I[1,] ±Aei1Aeik 0 = wWAe1 Ae = (detA)Card(W) e1e. This proves the identity wWdet(A-w)=Card(W)detA. The identity wWdet(1-Aw)=Card(W) is proved in exactly the same fashion. It remains to show that for all AEnd(V) there exists wW such that Aw does not have any nonzero fixed point. Suppose that there does not exist wW such that Aw has no nonzero fixed point. Then for each wW there is a nonzero vector vV such that (1-Aw)v=0. So, for each wW, det(1-Aw)=0. So wWdet(1-Aw)=0. This is a contradiction to the identity above.
    3. : Suppose that Γ is connected. Assume that V is a nonzero W-submodule of V. Let xV, x0. Then sixx for some i[1,]={1,2,,} since x,ei=0 for all 1i is impossible. Let i[1,] be such that sixx. Since si is a pseudo reflection we have that (1-si)x=ceiV. Let k be any other element of [1,]. Since Γ is connected there is a path ei=ei1ei2eim=ek in Γ connecting ei and ek. Assume eijV. Since sij+1 is a pseudoreflection, (1-sij+1)eij=ceij+1V. Since eij+1 and eij are connected in the graph Γ they are not orthogonal and thus sij+1eijeij and so c0. So eij+1V. In this way we show that ekV. Thus ekV for all k[1,]. So V=V. Thus V is a simple W-module.
      : Suppose that Γ is not connected and let Γ1 be a connected component of Γ. Let ei1,ei2,,eik be the vertices in Γ1. Then let V1 be the span of the basis elements eij corresponding to the vertices in Γ1. Let k[1,] and let eiV1 Then eiV1 and if ek is a vertex in Γ1, then, since (1-sk)ei=cekV1, we have that skeiV1. If ek is not a vertex in Γ1 then ek,ei=0 and it follows that skei=eiV1. Thus, V1 is stable under the action of W. So V1 is a proper submodule of V. So V is not simple.
    4. We will do this problem by completing the following steps.
      1. Let sjS. Then the rank of sj on pV is (n-1i-1).
      2. The modules iV and jV are nonisomorphic if ij.
      Let j be a vertex in the graph Γ such that Γ-{j} is connected. Let V=span {ei|ij} and let e(V) so that V=Ve.
      1. sje=e+cej, where c and c0.
      2. sje=v+de, where vV, v0 and d.
      It is clear that 0V and nV are irreducible since they are 1-dimensional. Let 0<p<n. Then pV=U1U2, where U1=pV and U2=(p-1V)e. By induction on dimV we know that U1 and U2 are irreducible under the action of W.
      1. U1 and U2 are not invariant under the action of W.
      2. If E is a nonzero submodule of pV then E=pV.
      1. Note that if v1,,vpV then, since sj is a pseudo-reflection we have that for every i, sjvi=vi+diej for some di. Thus (sj-1)v1 vp = sjv1sj vp-v1vp = (v1+d1ej) (vp+dpej)-v1 vp = k=1pdkv a1ej vp, and also sj(ejv1vp-1) = cjej (v1+d1ej) (vp-1+dp-1ej) -ejv1vp-1 = (cj-1)ej v1vp-1. The first equality shows that the action of sj on pV is a map onto ej(p-1V). The second equality shows that (sj-1) acts by a nonzero constant on ej(p-1V). It follows that the rank of sj-1 acting on pV is dim(ej(p-1V))=(n-1p-1).
      2. Assume that ij and that iV and jV are isomorphic. Then we have that (ni)=dim (iV)=dim (jV)= (nn-j), giving that i=n-j. We also must have that the rank of sj-1 on iV is the same as the rank of sj-1 on jV. It follows that (n-1i-1)=(n-1n-1-(j-1)), from which we get that i-1=n-1-j+1 or i=n-j+1. This is a contradiction to i=n-j. Thus iV and jV are nonisomorphic.
      3. Since sj is a pseudo-reflection, (1-sj)e=cej. Since ej is not in V we know that eNOTPERPej. So sjee. So c0.
      4. We know sje=v+de for some vV and some d. If v=0 then sje=de. In view of c) we see that this implies that ej=d-1ce, which is a contradiction since ejNOTPERPV. So v0.
      5. We have assumed that 0<p<n. So p-1n-2 and thus we can choose vectors v1,v2,,vp-1V such that vi is orthogonal to ej for all i. Let u0=v1vp-1. Then, by d), sj(u0e)= sju0sje= u0(v+de) =u0v+du0 e, where u0vU1 and is nonzero. This argument shows that U2 is not invariant under the action of W. Since U1=U2 in pV and we know that the inner product is W-invariant, we get also that U1 is not invariant under the action of W.
      6. Let EpV be a nonzero submodule. Suppose that EU1=0. Then let uE, u0, and write u=u1+u2 where uiUi. Since EU1=0, u20. By the induction hypothesis applied to W we know that U1 and U2 are irreducible and inequivalent. Thus there exists an element a[W] such that au1=0 and au20. So au=au2E. So EU20. Since U2 is irreducible as a W-module, U2E. Then, by d) we know that there is an element uE such that u=u1+u2 and u10. So u1=u2-uE. Again, by the irreducibility of U1 as a W-module, U1E. So E=pV. A similar argument holds in the case that EU2=0. Thus pV is irreducible.

    Notes. Bourbaki notes that this exercise is due to R. Steinberg. The solution to part d) is also due to R. Steinberg (unpublished notes).
    Concerning part b): Let F be a field of arbitrary characteristic, let G be a finite group, and let U be a finite dimensional G-module. For gG let gU denote the corresponding endomorphism of U. Let UG be the submodule of G-invariants. A projection onto UG means an idempotent F-linear transformation with range UG.

Lemma: Let π be any projection of U onto UG. Suppose bEnd(U) satisfies bUGUG. Then gGtrace(bgU) =|G|trace(bπ).

Corollary: If UG=0 and bEnd(U) then gGtrace(bgU)=0.

Proof of Lemma: Choose an F-subspace U of U such that U=UGU and use matrices relative to this decomposition. If aEnd(U) let [a] denote the corresponding matrix. Then |G|[π]= ( |G|IC12 00 ) gG[gU]= ( |G|ID12 00 ) where I is the identity matrix of size dimU and C12,D12 are matrices of appropriate size. Since bUGUG we have [b]= ( B11 B12 0 B22 ) . Thus |G|[b][π]= ( |G|B11I B11C12 0 0 ) and[b] gG[gU]= ( |G|B11I B11D12 0 0 ) . The assertion follows.

Remark: Note that the argument shows that [b](|g|[π]-gG[gU]) is nilpotent.

Now let G be a finite group and let V be a G-module. Assume as in the above exercise that (pV)G=0 for 1pn but make no hypothesis on the characteristic of the ground field F. We may view pV as a module for the semigroup End(V)G. If aEnd(V) let pa be the corresponding endomorphism of pV. Thus p(ab)=pa·pb. For aEnd(V) we have det(1-a)= p=0n (-1)ptrace (pa). Let’s take this as known. Replace a by ag. Then det(1-ag)= p=0n trace(pa·pg). If p>0, use the hypothesis (pV)G=0 and apply the Corollary with U=pV and b=pa. This gives gGtrace(pa·pg)=0 so gGdet(1-ag)= gGtrace (0a0g)= |G|.

page history