Fuetterer, Schwab - Spectral Sequences
Transcript of Fuetterer, Schwab - Spectral Sequences
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 119
Spectral Sequences
Miriam Schwab Michael Fuumlerer
ᵗʰ September
We explain the notion of a spectral sequence and various ways of construct-
ing them before illustrating the concept by some examples We do not treat
spectral sequences in the greatest possible generality but tried to keep the text
accessible
is note grew out of the seminar ldquoStable cohomology of the mapping classgrouprdquo by JProf Dr Gabriela Weitze-Schmithuumlsen and Dipl-Inform Tobias
Columbus at the Karlsruhe Institute of Technology in summer term We
would like to thank both of them and especially Tobias for explaining us a lot
about spectral sequences
Parts of this text follow [Wei] rather closely
Contents
1 Spectral Sequences and Convergence 2
2 Construction of Spectral Sequences 6
3 Double Complexes 10
4 Examples 12
e five lemma
e snake lemma
Balancing Tor Base change for Tor
e universal coefficients theorem
References 19
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 219
1 Spectral Sequences and Convergence
Let C be an abelian category
Definition (Spectral sequences) A (homologically graded) spectral sequence is a family
of objects E r pq
of C for all p q isin Z and r ge a (with a fixed a isin Z) together with
differentials d r pq E r pq
E r p minusr q +r minus1 which satisfy d r d r = 0 Furthermore we require
that there are isomorphisms
E r +1 pq H (E
r pq ) = ker(d r pq )
1048623im(d r p +r q minusr +1)
By n = p + q we denote the total degree of E r pq
e collections (E r pq ) p q isinZ for fixed r are called the sheets or pages of the spectral sequence
By the isomorphisms required above we imagine that we get from one sheet to the next
one (ldquoturing a page aroundrdquo) by taking homology
To get an imagination of how the differentials look like we first regard the sheet of thespectral sequence where r = 0
E 003 E 013 E 023 E 033
E 002 E 012 E 022 E 032
E 001 E 011 E 021 E 031
E 000 E 010 E 020 E 030
On the first sheet the differentials go from the right to the le as is shown here
E 103 E 113 E 123 E 133
E 1
02 E 1
12 E 1
22 E 1
32
E 101 E 111 E 121 E 131
E 100 E 110 E 120 E 130
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 319
In the following picture we see as an example the 2-sheet with differentials d 2 pq E 2 p q
E 2 p minus2q +1
E 203 E 213 E 223 E 233
E 202
E 212
E 222
E 232
E 201 E 211 E 221 E 231
E 200 E 210 E 220 E 230
Example (First quadrant spectral sequence) A spectral sequence with E r pq = 0 for p lt
0 or q lt 0 is called a first quadrant spectral sequence We observe that for r large enough
the differential with codomain E r pq has domain 0 and the differential with domain E r pq has
codomain 0 We get
E r +1 pq = H (E r pq ) = ker(d )
983087im(d ) = E r pq
9830870 = E r pq
e stable value E r pq = E k pq for k ge such r is named E infin pq
We restrict ourselves to first quadrant spectral sequences which makes our life much easier
Definition (Convergence) Let H n be a family of objects of C
We say a spectral sequence
(a) weakly converges to H lowast if there exists a filtration
sube F p minus1H n sube F p H n sube F p +1H n sube sube H n
for each n isin Z and furthermore isomorphisms
β pq E infin pq
F p H p +q 983087F p minus1H p +q
(b) approaches H lowast if it weakly converges to H lowast andsup1
H n =991786
F p H n and991785
F p H n = 0
(c) converges to H lowast if it approaches H lowast and H n = limlarrminusminus
983080H nF p H n
983081sup2
We denote convergence by E r pq rArr H p +q
Remark Let a spectral sequence converging to H lowast have E 2 pq = 0 unless p = 0 1 en
for each n there is a short exact sequence
0 E 20n H n E 21nminus1
0
sup1 e first of the two conditions is called exhaustive and the second one Hausdorff sup2 A filtration which satisfies foralln existt F t H n = H n fulfills this additional condition
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 419
Proof First we draw a picture of the 2-sheet of the spectral sequence
0
0
E 203
E 213
0
0
0
0
E 2
02
E 2
12
0
0
0 0 E 201 E 211 0 0
0
0 E 200 E 210
0
0
We see that E 2 pq = E infin pq and since we know that the spectral sequence converges we know
that there exists a filtration F p H n which fulfills E infin pq F p H p +q F p minus1H p +q
If p 0 1 we get 0 = E 2 pq = F p H p +q F p minusq H p +q which tells us F p H p +q = F p minus1H p +q and
therefore the filtration looks like
F minus2H n = F minus1H n sube F 0H n sube F 1H n = F 2H n =
Moreover we know thatcap
F p H n = 0 so F minus1H n = F minus2H n = = 0 and sincecup
F p H n =H n we get F 1H n = F 2H n = H n
Now let p = 0 We notice that
E 20n = E infin0n F 0H n
983087F minus1H n = F 0H n
For p = 1 we get
E 21nminus1 = E infin1nminus1 F 1H n 983087F 0H n = H n 983087F 0H n
and obviously the short exact sequence
0 F 0H n H n H n983087F 0H n
0
turns into the short exact sequence
0 E 20n H n E 21nminus1
0
Remark Let a spectral sequence converging to H lowast have E 2 pq = 0 unless q = 0 1 en
there is a long exact sequence
H p +1 E 2 p +10
d E 2 p minus11
H p E 2 p 0
d E 2 p minus21
H p minus1
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 519
Proof Again we first draw a picture
0
0
0
0
E 2minus11
E 2
01
E 2
11
E 2
21
E 2minus10 E 200 E 210 E 220
0
0
0
0
and we see that the objects at the 3-sheet are the objects E infin pq For q 1 0 we get
F p H p +q 983087F p minus1H p +q E
infin pq = E 3 pq = H (E 2 pq ) = ker(d 2 pq )
1048623im(d 2 p +2q minus1) = 0
and therefore F p H p +q = F p minus1H p +q
Letrsquos have a look at the filtration of H n
= F nminus3H n = F nminus2H n sube F nminus1H n sube F nH n = F n+1H n =
capF p H n = 0 implies F r H n = 0 for r le p +q minus 2 and
cupF p H n = H n implies F r H n = H n for
r ge n = p + q
Let q = 0 We get
H p 983087F p minus1H p = F p H p
983087F p minus1H p E
infin p 0 = E 3 p 0 = ker(d 2 p 0)
1048623im(d 2 p +2minus1) = ker(d 2 p 0)
And q = 1 leads to
F p H p +1 = F p H p +1983087F p minus1H p +1 E
infin p 1 = E 3 p 1
= H (E 2 p 1) = ker (d 2 p 1)1048623
im(d 2 p +20) = E 2 p 1
1048623im(d 2 p +20)
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 619
We can now put all this together to a long exact sequence
0
0
E 2 p +10 kerd 2 p +10
H p F p minus1H p
H p +1 E 2 p +10 E 2 p minus11 H p
kerd 2 p +10 F p minus1H p
0 0 0 0
2 Construction of Spectral Sequences
In this section we introduce the important notions of exact couples and filtered complexes
Every exact couple yields a spectral sequence and every filtered complex yields an exact
couple Finally we prove a convergence theorem for spectral sequences obtained in this
way
Construction (Exact couples and their derivations) An exact couple is a pair of objects
D and E of C and morphisms i j and k such that the diagram
D
D
E
i
j k
is exact at each vertex Since d = jk complies with d 2 = ( jk )2 = j (k j )k = j 0k = 0 we
can apply homology by seing H (E ) = ker(d )im(d ) and get the derived exact couple
i (D ) i (D )
H (E )
i |i (D )
j (2)k (2)
with j (2)(i (x )) = [ j (x )] and k (2)([e ]) = k (e ) x isin D and [e ] isin H (E ) It is an easy computa-
tion that j (2) and k (2) are well-defined and that the derived couple is exact It suggests itself
to iterate this process and set E 1 = E and E r = H (E r minus1) d 1 = d = jk and d r = j (r )k (r )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 719
e (r + 1)-th exact couple looks like
i r (D ) i r (D )
E r +1
i |i r (D )
j (r +1)k (r +1)
Remark Notice that the cycles can be described as Z r = k minus1(i r (D )) and the boundaries
as B r = j (ker(i r ))
Proof One just calculates
Z r = ker(d r ) = ker( j (r )k (r )) = ker( ji minusr +1k )
= k minus1(ker( ji minusr +1)) = k minus1(i r minus1(ker j ))
= k minus1(i r minus1(im i )) = k minus1(i r (D ))
B r = im(d r ) = im( j (r )k (r )) = im( ji minusr +1k )
= j (i minusr +1k (E pq )) = j (i minusr +1 im(k ))
= j (i minusr +1(ker(i )) = j (i minusr (0)) = j (ker(i r ))
We now regardD =oplus
D pq and E =oplus
E pq with the morphismsi j and k having bidegree
(1 minus1) (0 0)sup3 and (minus1 0) respectively (ie i (D pq ) sube D p +1q minus1 and so on) e bidegrees
of i and k do not change by passage to the derived couple while the bidegree of j (r ) in the
r -th couple is deg( j (r )) = deg( j ) minus (r minus 1) deg(i ) = (0 0) minus (r minus 1)(1 minus1) = (minus(r minus 1) r minus 1)since j (r )(i r minus1(x )) = [ j (x )] and therefore the bidegree of the differential d r = j (r )k (r ) is
deg(d r ) = (minusr r minus 1) just as claimed in the definition of spectral sequences
In this way the exact couple yields a spectral sequence with objects E
r
pq Example (Exact couple of a filtered complex) Let C lowast be a complex with a filtration
F p C lowast sube F p +1C lowast sube sube C lowast
such that there are integers s lt t for eachn with F s C n = 0 and F t C n = C n (Such a filtration
is called bounded Particularly this means F k C n = 0 for all k le s and F k C n = C n for all
k ge t We will always consider canonically bounded filtrations ie s = minus1 and F nC n = C nfor all n ndash what leads to a first quadrant spectral sequence)
We get short exact sequences
0 F p minus1C lowast
i
F p C lowast
π p
F p C lowast983087F p minus1C lowast
0
and by applying homology long exact secquences
H p minus1q +1(F p minus1C lowast)
i H p +q (F p C lowast)
j H p +q
983080F p C lowast
983087F p minus1C lowast
983081
δ H p minus1+q (F p C lowast)
sup3 To avoid confusion because of all the indices and variables we assume a = 0 ndash but j could also have bidegree(minusa a ) a 0 would not be more difficult but we would have to take care of it
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 819
where i and j are the maps induces by i and π p above and δ is the map delivered by the
snake lemma
Now we can roll up those long exact sequences into an exact triangle
oplusH p +q (F p C lowast) oplusH p +q (F p C lowast)
oplusH p +q (F p C lowastF p minus1C lowast)
i
j
k
which gives us a spectral sequence E r pq
From now on we concentrate on spectral sequences arising from an exact couple which
complies the following two conditions we require D pq = 0 if p lt 0 and for q lt 0 we want
i D p minus1q +1 D pq to be an isomorphism
Remark ose two facts guarantee that the spectral sequence lives in the first quadrant
Proof First let p lt 0 Since k E pq D p minus1q = 0 has im(k ) = 0 we get d = jk = 0
Furthermore 0 = D pq j
E pq says that
E pq = ker(d )983087
im(d ) = ker(d ) = im( j ) = 0
Now we look at the case q lt 0 We have d E pq D p minus1q
j
D p minus1q and we know
that i D p minus1q sim D p q minus1 is an isomorphism But since im(k ) = ker(i ) = 0 this leads to
d = jk = 0 Moreover we have D p minus1q +1
i
sim D pq j
E pq and since ker( j ) = im(i ) =D p minus1q +1 we get
E pq = ker(d ) = ker(k ) = im( j ) = 0
Let Z r pq = ker(d r pq ) and B r pq = im(d r p +r q minusr +1) be the cycles and boundaries given by the
fact that the r -sheet of the spectral sequence contains chain complexes
Furthermore we set H n = limminusminusrarr
D p nminus p (the injective limit along the maps i D pq
D p +1q minus1 ie the disjoint union of the D p nminus p where two elements are said to be equal if
they are equal under compositions of i ) and F p H n = im(D pq H p +q ) by what we get
a filtration
F p minus1H n sube F p H n sube
since i (D p minus1nminus p minus1) sube D p nminus p is filtration is even canonically bounded D p nminus p = 0for p lt 0 causes F p H n = 0 If p gt n (which is equal to q = n minus p lt 0) we know that
D p n minus p D p +1nminus( p +1) and therefore H p +q = im(D p nminus p H p +q ) = F p H n
In particular the filtration satisfiescup
p F p H n = H n = limlarrminusminus
(H n F p H n) andcap
p F p H n = 0
So if a spectral sequence weakly converges to H lowast it even converges to H lowast
Proposition ere is a natural inclusion of F p H n F p minus1H n in E infin p nminus p e spectral se-
quence E r pq weakly converges to H lowast if and only if
Z infin =991785r
k minus1(i r D ) equals k minus1(0) = j (D )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 919
Proof First we define K pq to be the kernel of D pq H p +q Beeing an element of K pq is
equivalent to lying in the kernel of i r for an integer r so K pq =cup
r ker(i r ) It follows that
j (K pq ) =cup
r j (ker(i r )) =cup
r B r pq = B infin pq as one can easily see that the infinite boundaries
are the union of the B r pq (and similarly Z infin pq =cap
r Z r pq ) We look at the following diagram
0
0 K p minus1q +1 D p minusq q +1 F p minus1H n
0
0 K pq D pq F p H n 0
coker(i |K p minus1q +1)
coker(i ) F p H nF p minus1H n
i |K p minus1q +1 i
with short exact sequences as rows
Before applying the snake lemma we regard the cokernels
D pq 983087i (D p minus1q +1) = D pq
983087im(i ) = D pq
983087ker( j )
tells us that coker(i ) = j (D pq ) Furthermore we have im(i |K p minus1q +1) = ker( j |K pq ) () and
therefore coker(i |K p minus1q +1) j (K pq ) = B infin pq e snake lemma now says that
0 B infin pq j (D pq ) F p H n
983087F p minus1H n
0
is a short exact sequence
For all r isin Z we have
j (D pq ) = im( j ) = ker(k ) = k minus1(0) sube k minus1(i r D p minusr minus1q +r ) = Z r pq
what means j (D pq ) subecap
r Z r pq = Z infin pq We get a natural inclusion
F p H n983087F p minusq H n
j (D pq )1048623B infin pq sube Z infin pq
1048623B infin pq = E infin pq
and in particular we know now that the spectral sequences weakly converges to H lowast ie
F p H n F p minus1H n = E infin pq if and only if j (D ) = Z infin
Corollary (Convergence eorem) A spectral sequence with the requirements mentioned
above converges to H lowast = limminusminusrarr D E r pq rArr H p +q
If a spectral sequence arises from an exact couple of a complex C lowast with canonically bounded
filtration it converges to limminusminusrarr
D = H lowast which is the homology of the filtered complex C lowast
Proof A remark above tells us that the spectral sequence converges to H lowast if it weakly con-
verges to H lowast All we have to show is that Z infin = k minus1(0) if the spectral sequence complies
the two facts that D pq = 0 for p lt 0 and i D p minus1q +1
sim D pq for q lt 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1019
Let p and q be arbitrary integers ere exists an integer r such that p minus r minus 1 lt 0 which
implies i r (D p minusr minus1q +r ) = i r (0) = 0 We now see that Z r pq = k minus1(i r (D p minusr minus1q +r ) =
k minus1(0) en Z infin pq =cap
r prime Z r prime
pq = k minus1(0) and therefore Z infin = k minus1(0)
at means E r pq rArr H p +q
Let F p minus1C lowast sube F p C lowast sube sube C lowast be a canonically bounded filtration of a complex C lowastSince F p C n = 0 for all p lt 0 we get D pq = H p +q (F p C lowast) = 0 for those p For q lt 0 we have
p = n minusq gt n and F p C n = C n is leads to D pq = H n(F p C lowast) = H n(C lowast) = H n(F p +1C lowast) =D p +1q minus1 = D p +1q minus2 Alltogether it follows that a spectral sequence arising from a
canonically bounded complex converges to H lowast
It remains to show hat H lowast is the homology of C lowast
Let p lt n we then know that i is an isomorphism and therefore we get
H p +q im(D p nminus p H p +q ) D p nminus p = H p +q (F p C lowast) = H p +q (C lowast)
3 Double ComplexesIn Example we explained that a filtered complex yields an exact couple which in turn
yields a spectral sequence We will now go one step further and introduce double complexes
which yield filtered complexes and finally spectral sequences
Definition (Double complex) A double complex C lowastlowast is a family (C pq ) p q isinZ of objects
of an abelian category together with maps
d h pq C pq C p minus1q d v pq C pq C p q minus1
such that for each q resp p the families (C lowastq d hlowastq ) resp (C p lowast d v p lowast) are chain complexes
and additionally in every square the relation
d v p minus1q d h pq = plusmn d h p q minus1 d v pq
is satisfied In case of a ldquo+rdquo in this relation we call the double complex commutative and
in case of a ldquominusrdquo we call it anticommutative We say that a double complex lives in the first
quadrant if C pq = 0 whenever p lt 0 or q lt 0
We can picture a double complex like this (here we have a first quadrant double complex)
C 02
C 12
C 22
C 01 C 11 C 21
C 00 C 10 C 20
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1119
Definition Let C lowastlowast be an anticommutative double complex en we define the total
complex aached to C lowastlowast as the chain complex Totlowast(C lowastlowast) with entries⁴
Totn(C lowastlowast) =991893
p +q =n
C pq
and differentials d = d v + d h
A straightforward calculation using anticommutativity shows d 2 = 0 so that the total
complex is indeed a chain complex
We can turn the total complex of a given double complex into a filtered chain complex by
considering the following two obvious filtrations
ldquoFiltration by rowsrdquo
0 0 0
lowast lowast lowast
lowast lowast lowast
ldquoFiltration by columnsrdquo
lowast lowast 0lowast lowast 0lowast lowast 0
More precisely we set
F (R) p Totn(C lowastlowast) =
991893i le p
C i nminusi F (C)q Totn(C lowastlowast) =
991893 j leq
C nminus j j
In both cases we get
983080F p Totlowast(C lowastlowast)
983087F p minus1 Totlowast(C lowastlowast)
983081n
= C pq
for the n-th entry of the filtration quotients (where p + q = n) Note that when the double
complex lives in the first quadrant then both filtrations are canonically bounded ereforewe will assume from now on that every double complex lives in the first quadrant
e machinery developped in the previous sections now yields two spectral sequences
which we call (R)E r lowastlowast and (C)E r lowastlowast respectively By Corollary both spectral sequences
converge to the homology of the total complex Totlowast(C lowastlowast) Oen one is not really inter-
ested in this homology but one can play these sequences off against one another to obtain
information about the entries of the spectral sequence
Can we write down more explicitly how these spectral sequences look like We have to
apply the technique of exact couples to our situation Doing so we obtain
Proposition e E 1 page of (C)E r lowastlowast is obtained from the initial double complex C lowastlowast by
taking homology in vertical direction and the differentials d 1 on that page are the maps in-
duced by the horizontal differentials d h of C lowastlowast on this homology Dually the E 1 page of (R)E r lowastlowastis obtained by ldquotransposingrdquo this procedure
⁴ It is also common to consider a slightly differently defined total complex in which the direct sum is replacedby a direct product As we will mostly consider double complexes living in the first quadrant this does notmake a difference for us
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1219
Idea of proof To prove this we would have to go into the details of how the long homology
sequence coming from the short exact sequence of chain complexes
0 F p minus1 Totlowast(C lowastlowast) F p Totlowast(C lowastlowast) F p Totlowast(C lowastlowast)1048623F p minus1 Totlowast(C lowastlowast) 0
is constructed which in the end boils down to investigating the definition of the connectinghomomorphism coming from the snake lemma As this is a rather tedious calculation we
will omit this here
is proposition suggests to view the initial double complex C lowastlowast as the E 0 page of either
spectral sequence (possibly ldquotransposedrdquo)
Before giving some examples of applications of this construction we record the following
useful result
Lemma If we have a converging spectral sequence
E r pq rArr H p +q
and the E 2 page consists of 0 everywhere except in the boom row then we have
E 2 p 0 = H p
Proof is follows directly from Remark
Finally we note that we can also apply this whole technique to a commutative double com-
plex We can turn any such double complex into an anticommutative one by inserting some
minus1rsquos for example at every second horizontal or vertical arrow in this way in every square
exactly one map is multiplied by minus1 is does not affect the remaining properties of being
a double complex and does also not change homology
4 Examples
We conclude this note with a few examples that demonstrate the power of our techniques
e first examples are very simple whereas the later ones are a bit more interesting
41 The five lemma
As a first example we want to prove the five lemma Given have a commutative diagram
A B C D E
Aprime B prime C prime D prime E prime
f
д
f prime
д prime
α β γ δ ε
with exact rows we want to prove
(a) When β δ are monomorphisms and α is an epimorphism then γ is a monomorphism
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 219
1 Spectral Sequences and Convergence
Let C be an abelian category
Definition (Spectral sequences) A (homologically graded) spectral sequence is a family
of objects E r pq
of C for all p q isin Z and r ge a (with a fixed a isin Z) together with
differentials d r pq E r pq
E r p minusr q +r minus1 which satisfy d r d r = 0 Furthermore we require
that there are isomorphisms
E r +1 pq H (E
r pq ) = ker(d r pq )
1048623im(d r p +r q minusr +1)
By n = p + q we denote the total degree of E r pq
e collections (E r pq ) p q isinZ for fixed r are called the sheets or pages of the spectral sequence
By the isomorphisms required above we imagine that we get from one sheet to the next
one (ldquoturing a page aroundrdquo) by taking homology
To get an imagination of how the differentials look like we first regard the sheet of thespectral sequence where r = 0
E 003 E 013 E 023 E 033
E 002 E 012 E 022 E 032
E 001 E 011 E 021 E 031
E 000 E 010 E 020 E 030
On the first sheet the differentials go from the right to the le as is shown here
E 103 E 113 E 123 E 133
E 1
02 E 1
12 E 1
22 E 1
32
E 101 E 111 E 121 E 131
E 100 E 110 E 120 E 130
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 319
In the following picture we see as an example the 2-sheet with differentials d 2 pq E 2 p q
E 2 p minus2q +1
E 203 E 213 E 223 E 233
E 202
E 212
E 222
E 232
E 201 E 211 E 221 E 231
E 200 E 210 E 220 E 230
Example (First quadrant spectral sequence) A spectral sequence with E r pq = 0 for p lt
0 or q lt 0 is called a first quadrant spectral sequence We observe that for r large enough
the differential with codomain E r pq has domain 0 and the differential with domain E r pq has
codomain 0 We get
E r +1 pq = H (E r pq ) = ker(d )
983087im(d ) = E r pq
9830870 = E r pq
e stable value E r pq = E k pq for k ge such r is named E infin pq
We restrict ourselves to first quadrant spectral sequences which makes our life much easier
Definition (Convergence) Let H n be a family of objects of C
We say a spectral sequence
(a) weakly converges to H lowast if there exists a filtration
sube F p minus1H n sube F p H n sube F p +1H n sube sube H n
for each n isin Z and furthermore isomorphisms
β pq E infin pq
F p H p +q 983087F p minus1H p +q
(b) approaches H lowast if it weakly converges to H lowast andsup1
H n =991786
F p H n and991785
F p H n = 0
(c) converges to H lowast if it approaches H lowast and H n = limlarrminusminus
983080H nF p H n
983081sup2
We denote convergence by E r pq rArr H p +q
Remark Let a spectral sequence converging to H lowast have E 2 pq = 0 unless p = 0 1 en
for each n there is a short exact sequence
0 E 20n H n E 21nminus1
0
sup1 e first of the two conditions is called exhaustive and the second one Hausdorff sup2 A filtration which satisfies foralln existt F t H n = H n fulfills this additional condition
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 419
Proof First we draw a picture of the 2-sheet of the spectral sequence
0
0
E 203
E 213
0
0
0
0
E 2
02
E 2
12
0
0
0 0 E 201 E 211 0 0
0
0 E 200 E 210
0
0
We see that E 2 pq = E infin pq and since we know that the spectral sequence converges we know
that there exists a filtration F p H n which fulfills E infin pq F p H p +q F p minus1H p +q
If p 0 1 we get 0 = E 2 pq = F p H p +q F p minusq H p +q which tells us F p H p +q = F p minus1H p +q and
therefore the filtration looks like
F minus2H n = F minus1H n sube F 0H n sube F 1H n = F 2H n =
Moreover we know thatcap
F p H n = 0 so F minus1H n = F minus2H n = = 0 and sincecup
F p H n =H n we get F 1H n = F 2H n = H n
Now let p = 0 We notice that
E 20n = E infin0n F 0H n
983087F minus1H n = F 0H n
For p = 1 we get
E 21nminus1 = E infin1nminus1 F 1H n 983087F 0H n = H n 983087F 0H n
and obviously the short exact sequence
0 F 0H n H n H n983087F 0H n
0
turns into the short exact sequence
0 E 20n H n E 21nminus1
0
Remark Let a spectral sequence converging to H lowast have E 2 pq = 0 unless q = 0 1 en
there is a long exact sequence
H p +1 E 2 p +10
d E 2 p minus11
H p E 2 p 0
d E 2 p minus21
H p minus1
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 519
Proof Again we first draw a picture
0
0
0
0
E 2minus11
E 2
01
E 2
11
E 2
21
E 2minus10 E 200 E 210 E 220
0
0
0
0
and we see that the objects at the 3-sheet are the objects E infin pq For q 1 0 we get
F p H p +q 983087F p minus1H p +q E
infin pq = E 3 pq = H (E 2 pq ) = ker(d 2 pq )
1048623im(d 2 p +2q minus1) = 0
and therefore F p H p +q = F p minus1H p +q
Letrsquos have a look at the filtration of H n
= F nminus3H n = F nminus2H n sube F nminus1H n sube F nH n = F n+1H n =
capF p H n = 0 implies F r H n = 0 for r le p +q minus 2 and
cupF p H n = H n implies F r H n = H n for
r ge n = p + q
Let q = 0 We get
H p 983087F p minus1H p = F p H p
983087F p minus1H p E
infin p 0 = E 3 p 0 = ker(d 2 p 0)
1048623im(d 2 p +2minus1) = ker(d 2 p 0)
And q = 1 leads to
F p H p +1 = F p H p +1983087F p minus1H p +1 E
infin p 1 = E 3 p 1
= H (E 2 p 1) = ker (d 2 p 1)1048623
im(d 2 p +20) = E 2 p 1
1048623im(d 2 p +20)
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 619
We can now put all this together to a long exact sequence
0
0
E 2 p +10 kerd 2 p +10
H p F p minus1H p
H p +1 E 2 p +10 E 2 p minus11 H p
kerd 2 p +10 F p minus1H p
0 0 0 0
2 Construction of Spectral Sequences
In this section we introduce the important notions of exact couples and filtered complexes
Every exact couple yields a spectral sequence and every filtered complex yields an exact
couple Finally we prove a convergence theorem for spectral sequences obtained in this
way
Construction (Exact couples and their derivations) An exact couple is a pair of objects
D and E of C and morphisms i j and k such that the diagram
D
D
E
i
j k
is exact at each vertex Since d = jk complies with d 2 = ( jk )2 = j (k j )k = j 0k = 0 we
can apply homology by seing H (E ) = ker(d )im(d ) and get the derived exact couple
i (D ) i (D )
H (E )
i |i (D )
j (2)k (2)
with j (2)(i (x )) = [ j (x )] and k (2)([e ]) = k (e ) x isin D and [e ] isin H (E ) It is an easy computa-
tion that j (2) and k (2) are well-defined and that the derived couple is exact It suggests itself
to iterate this process and set E 1 = E and E r = H (E r minus1) d 1 = d = jk and d r = j (r )k (r )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 719
e (r + 1)-th exact couple looks like
i r (D ) i r (D )
E r +1
i |i r (D )
j (r +1)k (r +1)
Remark Notice that the cycles can be described as Z r = k minus1(i r (D )) and the boundaries
as B r = j (ker(i r ))
Proof One just calculates
Z r = ker(d r ) = ker( j (r )k (r )) = ker( ji minusr +1k )
= k minus1(ker( ji minusr +1)) = k minus1(i r minus1(ker j ))
= k minus1(i r minus1(im i )) = k minus1(i r (D ))
B r = im(d r ) = im( j (r )k (r )) = im( ji minusr +1k )
= j (i minusr +1k (E pq )) = j (i minusr +1 im(k ))
= j (i minusr +1(ker(i )) = j (i minusr (0)) = j (ker(i r ))
We now regardD =oplus
D pq and E =oplus
E pq with the morphismsi j and k having bidegree
(1 minus1) (0 0)sup3 and (minus1 0) respectively (ie i (D pq ) sube D p +1q minus1 and so on) e bidegrees
of i and k do not change by passage to the derived couple while the bidegree of j (r ) in the
r -th couple is deg( j (r )) = deg( j ) minus (r minus 1) deg(i ) = (0 0) minus (r minus 1)(1 minus1) = (minus(r minus 1) r minus 1)since j (r )(i r minus1(x )) = [ j (x )] and therefore the bidegree of the differential d r = j (r )k (r ) is
deg(d r ) = (minusr r minus 1) just as claimed in the definition of spectral sequences
In this way the exact couple yields a spectral sequence with objects E
r
pq Example (Exact couple of a filtered complex) Let C lowast be a complex with a filtration
F p C lowast sube F p +1C lowast sube sube C lowast
such that there are integers s lt t for eachn with F s C n = 0 and F t C n = C n (Such a filtration
is called bounded Particularly this means F k C n = 0 for all k le s and F k C n = C n for all
k ge t We will always consider canonically bounded filtrations ie s = minus1 and F nC n = C nfor all n ndash what leads to a first quadrant spectral sequence)
We get short exact sequences
0 F p minus1C lowast
i
F p C lowast
π p
F p C lowast983087F p minus1C lowast
0
and by applying homology long exact secquences
H p minus1q +1(F p minus1C lowast)
i H p +q (F p C lowast)
j H p +q
983080F p C lowast
983087F p minus1C lowast
983081
δ H p minus1+q (F p C lowast)
sup3 To avoid confusion because of all the indices and variables we assume a = 0 ndash but j could also have bidegree(minusa a ) a 0 would not be more difficult but we would have to take care of it
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 819
where i and j are the maps induces by i and π p above and δ is the map delivered by the
snake lemma
Now we can roll up those long exact sequences into an exact triangle
oplusH p +q (F p C lowast) oplusH p +q (F p C lowast)
oplusH p +q (F p C lowastF p minus1C lowast)
i
j
k
which gives us a spectral sequence E r pq
From now on we concentrate on spectral sequences arising from an exact couple which
complies the following two conditions we require D pq = 0 if p lt 0 and for q lt 0 we want
i D p minus1q +1 D pq to be an isomorphism
Remark ose two facts guarantee that the spectral sequence lives in the first quadrant
Proof First let p lt 0 Since k E pq D p minus1q = 0 has im(k ) = 0 we get d = jk = 0
Furthermore 0 = D pq j
E pq says that
E pq = ker(d )983087
im(d ) = ker(d ) = im( j ) = 0
Now we look at the case q lt 0 We have d E pq D p minus1q
j
D p minus1q and we know
that i D p minus1q sim D p q minus1 is an isomorphism But since im(k ) = ker(i ) = 0 this leads to
d = jk = 0 Moreover we have D p minus1q +1
i
sim D pq j
E pq and since ker( j ) = im(i ) =D p minus1q +1 we get
E pq = ker(d ) = ker(k ) = im( j ) = 0
Let Z r pq = ker(d r pq ) and B r pq = im(d r p +r q minusr +1) be the cycles and boundaries given by the
fact that the r -sheet of the spectral sequence contains chain complexes
Furthermore we set H n = limminusminusrarr
D p nminus p (the injective limit along the maps i D pq
D p +1q minus1 ie the disjoint union of the D p nminus p where two elements are said to be equal if
they are equal under compositions of i ) and F p H n = im(D pq H p +q ) by what we get
a filtration
F p minus1H n sube F p H n sube
since i (D p minus1nminus p minus1) sube D p nminus p is filtration is even canonically bounded D p nminus p = 0for p lt 0 causes F p H n = 0 If p gt n (which is equal to q = n minus p lt 0) we know that
D p n minus p D p +1nminus( p +1) and therefore H p +q = im(D p nminus p H p +q ) = F p H n
In particular the filtration satisfiescup
p F p H n = H n = limlarrminusminus
(H n F p H n) andcap
p F p H n = 0
So if a spectral sequence weakly converges to H lowast it even converges to H lowast
Proposition ere is a natural inclusion of F p H n F p minus1H n in E infin p nminus p e spectral se-
quence E r pq weakly converges to H lowast if and only if
Z infin =991785r
k minus1(i r D ) equals k minus1(0) = j (D )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 919
Proof First we define K pq to be the kernel of D pq H p +q Beeing an element of K pq is
equivalent to lying in the kernel of i r for an integer r so K pq =cup
r ker(i r ) It follows that
j (K pq ) =cup
r j (ker(i r )) =cup
r B r pq = B infin pq as one can easily see that the infinite boundaries
are the union of the B r pq (and similarly Z infin pq =cap
r Z r pq ) We look at the following diagram
0
0 K p minus1q +1 D p minusq q +1 F p minus1H n
0
0 K pq D pq F p H n 0
coker(i |K p minus1q +1)
coker(i ) F p H nF p minus1H n
i |K p minus1q +1 i
with short exact sequences as rows
Before applying the snake lemma we regard the cokernels
D pq 983087i (D p minus1q +1) = D pq
983087im(i ) = D pq
983087ker( j )
tells us that coker(i ) = j (D pq ) Furthermore we have im(i |K p minus1q +1) = ker( j |K pq ) () and
therefore coker(i |K p minus1q +1) j (K pq ) = B infin pq e snake lemma now says that
0 B infin pq j (D pq ) F p H n
983087F p minus1H n
0
is a short exact sequence
For all r isin Z we have
j (D pq ) = im( j ) = ker(k ) = k minus1(0) sube k minus1(i r D p minusr minus1q +r ) = Z r pq
what means j (D pq ) subecap
r Z r pq = Z infin pq We get a natural inclusion
F p H n983087F p minusq H n
j (D pq )1048623B infin pq sube Z infin pq
1048623B infin pq = E infin pq
and in particular we know now that the spectral sequences weakly converges to H lowast ie
F p H n F p minus1H n = E infin pq if and only if j (D ) = Z infin
Corollary (Convergence eorem) A spectral sequence with the requirements mentioned
above converges to H lowast = limminusminusrarr D E r pq rArr H p +q
If a spectral sequence arises from an exact couple of a complex C lowast with canonically bounded
filtration it converges to limminusminusrarr
D = H lowast which is the homology of the filtered complex C lowast
Proof A remark above tells us that the spectral sequence converges to H lowast if it weakly con-
verges to H lowast All we have to show is that Z infin = k minus1(0) if the spectral sequence complies
the two facts that D pq = 0 for p lt 0 and i D p minus1q +1
sim D pq for q lt 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1019
Let p and q be arbitrary integers ere exists an integer r such that p minus r minus 1 lt 0 which
implies i r (D p minusr minus1q +r ) = i r (0) = 0 We now see that Z r pq = k minus1(i r (D p minusr minus1q +r ) =
k minus1(0) en Z infin pq =cap
r prime Z r prime
pq = k minus1(0) and therefore Z infin = k minus1(0)
at means E r pq rArr H p +q
Let F p minus1C lowast sube F p C lowast sube sube C lowast be a canonically bounded filtration of a complex C lowastSince F p C n = 0 for all p lt 0 we get D pq = H p +q (F p C lowast) = 0 for those p For q lt 0 we have
p = n minusq gt n and F p C n = C n is leads to D pq = H n(F p C lowast) = H n(C lowast) = H n(F p +1C lowast) =D p +1q minus1 = D p +1q minus2 Alltogether it follows that a spectral sequence arising from a
canonically bounded complex converges to H lowast
It remains to show hat H lowast is the homology of C lowast
Let p lt n we then know that i is an isomorphism and therefore we get
H p +q im(D p nminus p H p +q ) D p nminus p = H p +q (F p C lowast) = H p +q (C lowast)
3 Double ComplexesIn Example we explained that a filtered complex yields an exact couple which in turn
yields a spectral sequence We will now go one step further and introduce double complexes
which yield filtered complexes and finally spectral sequences
Definition (Double complex) A double complex C lowastlowast is a family (C pq ) p q isinZ of objects
of an abelian category together with maps
d h pq C pq C p minus1q d v pq C pq C p q minus1
such that for each q resp p the families (C lowastq d hlowastq ) resp (C p lowast d v p lowast) are chain complexes
and additionally in every square the relation
d v p minus1q d h pq = plusmn d h p q minus1 d v pq
is satisfied In case of a ldquo+rdquo in this relation we call the double complex commutative and
in case of a ldquominusrdquo we call it anticommutative We say that a double complex lives in the first
quadrant if C pq = 0 whenever p lt 0 or q lt 0
We can picture a double complex like this (here we have a first quadrant double complex)
C 02
C 12
C 22
C 01 C 11 C 21
C 00 C 10 C 20
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1119
Definition Let C lowastlowast be an anticommutative double complex en we define the total
complex aached to C lowastlowast as the chain complex Totlowast(C lowastlowast) with entries⁴
Totn(C lowastlowast) =991893
p +q =n
C pq
and differentials d = d v + d h
A straightforward calculation using anticommutativity shows d 2 = 0 so that the total
complex is indeed a chain complex
We can turn the total complex of a given double complex into a filtered chain complex by
considering the following two obvious filtrations
ldquoFiltration by rowsrdquo
0 0 0
lowast lowast lowast
lowast lowast lowast
ldquoFiltration by columnsrdquo
lowast lowast 0lowast lowast 0lowast lowast 0
More precisely we set
F (R) p Totn(C lowastlowast) =
991893i le p
C i nminusi F (C)q Totn(C lowastlowast) =
991893 j leq
C nminus j j
In both cases we get
983080F p Totlowast(C lowastlowast)
983087F p minus1 Totlowast(C lowastlowast)
983081n
= C pq
for the n-th entry of the filtration quotients (where p + q = n) Note that when the double
complex lives in the first quadrant then both filtrations are canonically bounded ereforewe will assume from now on that every double complex lives in the first quadrant
e machinery developped in the previous sections now yields two spectral sequences
which we call (R)E r lowastlowast and (C)E r lowastlowast respectively By Corollary both spectral sequences
converge to the homology of the total complex Totlowast(C lowastlowast) Oen one is not really inter-
ested in this homology but one can play these sequences off against one another to obtain
information about the entries of the spectral sequence
Can we write down more explicitly how these spectral sequences look like We have to
apply the technique of exact couples to our situation Doing so we obtain
Proposition e E 1 page of (C)E r lowastlowast is obtained from the initial double complex C lowastlowast by
taking homology in vertical direction and the differentials d 1 on that page are the maps in-
duced by the horizontal differentials d h of C lowastlowast on this homology Dually the E 1 page of (R)E r lowastlowastis obtained by ldquotransposingrdquo this procedure
⁴ It is also common to consider a slightly differently defined total complex in which the direct sum is replacedby a direct product As we will mostly consider double complexes living in the first quadrant this does notmake a difference for us
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1219
Idea of proof To prove this we would have to go into the details of how the long homology
sequence coming from the short exact sequence of chain complexes
0 F p minus1 Totlowast(C lowastlowast) F p Totlowast(C lowastlowast) F p Totlowast(C lowastlowast)1048623F p minus1 Totlowast(C lowastlowast) 0
is constructed which in the end boils down to investigating the definition of the connectinghomomorphism coming from the snake lemma As this is a rather tedious calculation we
will omit this here
is proposition suggests to view the initial double complex C lowastlowast as the E 0 page of either
spectral sequence (possibly ldquotransposedrdquo)
Before giving some examples of applications of this construction we record the following
useful result
Lemma If we have a converging spectral sequence
E r pq rArr H p +q
and the E 2 page consists of 0 everywhere except in the boom row then we have
E 2 p 0 = H p
Proof is follows directly from Remark
Finally we note that we can also apply this whole technique to a commutative double com-
plex We can turn any such double complex into an anticommutative one by inserting some
minus1rsquos for example at every second horizontal or vertical arrow in this way in every square
exactly one map is multiplied by minus1 is does not affect the remaining properties of being
a double complex and does also not change homology
4 Examples
We conclude this note with a few examples that demonstrate the power of our techniques
e first examples are very simple whereas the later ones are a bit more interesting
41 The five lemma
As a first example we want to prove the five lemma Given have a commutative diagram
A B C D E
Aprime B prime C prime D prime E prime
f
д
f prime
д prime
α β γ δ ε
with exact rows we want to prove
(a) When β δ are monomorphisms and α is an epimorphism then γ is a monomorphism
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 319
In the following picture we see as an example the 2-sheet with differentials d 2 pq E 2 p q
E 2 p minus2q +1
E 203 E 213 E 223 E 233
E 202
E 212
E 222
E 232
E 201 E 211 E 221 E 231
E 200 E 210 E 220 E 230
Example (First quadrant spectral sequence) A spectral sequence with E r pq = 0 for p lt
0 or q lt 0 is called a first quadrant spectral sequence We observe that for r large enough
the differential with codomain E r pq has domain 0 and the differential with domain E r pq has
codomain 0 We get
E r +1 pq = H (E r pq ) = ker(d )
983087im(d ) = E r pq
9830870 = E r pq
e stable value E r pq = E k pq for k ge such r is named E infin pq
We restrict ourselves to first quadrant spectral sequences which makes our life much easier
Definition (Convergence) Let H n be a family of objects of C
We say a spectral sequence
(a) weakly converges to H lowast if there exists a filtration
sube F p minus1H n sube F p H n sube F p +1H n sube sube H n
for each n isin Z and furthermore isomorphisms
β pq E infin pq
F p H p +q 983087F p minus1H p +q
(b) approaches H lowast if it weakly converges to H lowast andsup1
H n =991786
F p H n and991785
F p H n = 0
(c) converges to H lowast if it approaches H lowast and H n = limlarrminusminus
983080H nF p H n
983081sup2
We denote convergence by E r pq rArr H p +q
Remark Let a spectral sequence converging to H lowast have E 2 pq = 0 unless p = 0 1 en
for each n there is a short exact sequence
0 E 20n H n E 21nminus1
0
sup1 e first of the two conditions is called exhaustive and the second one Hausdorff sup2 A filtration which satisfies foralln existt F t H n = H n fulfills this additional condition
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 419
Proof First we draw a picture of the 2-sheet of the spectral sequence
0
0
E 203
E 213
0
0
0
0
E 2
02
E 2
12
0
0
0 0 E 201 E 211 0 0
0
0 E 200 E 210
0
0
We see that E 2 pq = E infin pq and since we know that the spectral sequence converges we know
that there exists a filtration F p H n which fulfills E infin pq F p H p +q F p minus1H p +q
If p 0 1 we get 0 = E 2 pq = F p H p +q F p minusq H p +q which tells us F p H p +q = F p minus1H p +q and
therefore the filtration looks like
F minus2H n = F minus1H n sube F 0H n sube F 1H n = F 2H n =
Moreover we know thatcap
F p H n = 0 so F minus1H n = F minus2H n = = 0 and sincecup
F p H n =H n we get F 1H n = F 2H n = H n
Now let p = 0 We notice that
E 20n = E infin0n F 0H n
983087F minus1H n = F 0H n
For p = 1 we get
E 21nminus1 = E infin1nminus1 F 1H n 983087F 0H n = H n 983087F 0H n
and obviously the short exact sequence
0 F 0H n H n H n983087F 0H n
0
turns into the short exact sequence
0 E 20n H n E 21nminus1
0
Remark Let a spectral sequence converging to H lowast have E 2 pq = 0 unless q = 0 1 en
there is a long exact sequence
H p +1 E 2 p +10
d E 2 p minus11
H p E 2 p 0
d E 2 p minus21
H p minus1
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 519
Proof Again we first draw a picture
0
0
0
0
E 2minus11
E 2
01
E 2
11
E 2
21
E 2minus10 E 200 E 210 E 220
0
0
0
0
and we see that the objects at the 3-sheet are the objects E infin pq For q 1 0 we get
F p H p +q 983087F p minus1H p +q E
infin pq = E 3 pq = H (E 2 pq ) = ker(d 2 pq )
1048623im(d 2 p +2q minus1) = 0
and therefore F p H p +q = F p minus1H p +q
Letrsquos have a look at the filtration of H n
= F nminus3H n = F nminus2H n sube F nminus1H n sube F nH n = F n+1H n =
capF p H n = 0 implies F r H n = 0 for r le p +q minus 2 and
cupF p H n = H n implies F r H n = H n for
r ge n = p + q
Let q = 0 We get
H p 983087F p minus1H p = F p H p
983087F p minus1H p E
infin p 0 = E 3 p 0 = ker(d 2 p 0)
1048623im(d 2 p +2minus1) = ker(d 2 p 0)
And q = 1 leads to
F p H p +1 = F p H p +1983087F p minus1H p +1 E
infin p 1 = E 3 p 1
= H (E 2 p 1) = ker (d 2 p 1)1048623
im(d 2 p +20) = E 2 p 1
1048623im(d 2 p +20)
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 619
We can now put all this together to a long exact sequence
0
0
E 2 p +10 kerd 2 p +10
H p F p minus1H p
H p +1 E 2 p +10 E 2 p minus11 H p
kerd 2 p +10 F p minus1H p
0 0 0 0
2 Construction of Spectral Sequences
In this section we introduce the important notions of exact couples and filtered complexes
Every exact couple yields a spectral sequence and every filtered complex yields an exact
couple Finally we prove a convergence theorem for spectral sequences obtained in this
way
Construction (Exact couples and their derivations) An exact couple is a pair of objects
D and E of C and morphisms i j and k such that the diagram
D
D
E
i
j k
is exact at each vertex Since d = jk complies with d 2 = ( jk )2 = j (k j )k = j 0k = 0 we
can apply homology by seing H (E ) = ker(d )im(d ) and get the derived exact couple
i (D ) i (D )
H (E )
i |i (D )
j (2)k (2)
with j (2)(i (x )) = [ j (x )] and k (2)([e ]) = k (e ) x isin D and [e ] isin H (E ) It is an easy computa-
tion that j (2) and k (2) are well-defined and that the derived couple is exact It suggests itself
to iterate this process and set E 1 = E and E r = H (E r minus1) d 1 = d = jk and d r = j (r )k (r )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 719
e (r + 1)-th exact couple looks like
i r (D ) i r (D )
E r +1
i |i r (D )
j (r +1)k (r +1)
Remark Notice that the cycles can be described as Z r = k minus1(i r (D )) and the boundaries
as B r = j (ker(i r ))
Proof One just calculates
Z r = ker(d r ) = ker( j (r )k (r )) = ker( ji minusr +1k )
= k minus1(ker( ji minusr +1)) = k minus1(i r minus1(ker j ))
= k minus1(i r minus1(im i )) = k minus1(i r (D ))
B r = im(d r ) = im( j (r )k (r )) = im( ji minusr +1k )
= j (i minusr +1k (E pq )) = j (i minusr +1 im(k ))
= j (i minusr +1(ker(i )) = j (i minusr (0)) = j (ker(i r ))
We now regardD =oplus
D pq and E =oplus
E pq with the morphismsi j and k having bidegree
(1 minus1) (0 0)sup3 and (minus1 0) respectively (ie i (D pq ) sube D p +1q minus1 and so on) e bidegrees
of i and k do not change by passage to the derived couple while the bidegree of j (r ) in the
r -th couple is deg( j (r )) = deg( j ) minus (r minus 1) deg(i ) = (0 0) minus (r minus 1)(1 minus1) = (minus(r minus 1) r minus 1)since j (r )(i r minus1(x )) = [ j (x )] and therefore the bidegree of the differential d r = j (r )k (r ) is
deg(d r ) = (minusr r minus 1) just as claimed in the definition of spectral sequences
In this way the exact couple yields a spectral sequence with objects E
r
pq Example (Exact couple of a filtered complex) Let C lowast be a complex with a filtration
F p C lowast sube F p +1C lowast sube sube C lowast
such that there are integers s lt t for eachn with F s C n = 0 and F t C n = C n (Such a filtration
is called bounded Particularly this means F k C n = 0 for all k le s and F k C n = C n for all
k ge t We will always consider canonically bounded filtrations ie s = minus1 and F nC n = C nfor all n ndash what leads to a first quadrant spectral sequence)
We get short exact sequences
0 F p minus1C lowast
i
F p C lowast
π p
F p C lowast983087F p minus1C lowast
0
and by applying homology long exact secquences
H p minus1q +1(F p minus1C lowast)
i H p +q (F p C lowast)
j H p +q
983080F p C lowast
983087F p minus1C lowast
983081
δ H p minus1+q (F p C lowast)
sup3 To avoid confusion because of all the indices and variables we assume a = 0 ndash but j could also have bidegree(minusa a ) a 0 would not be more difficult but we would have to take care of it
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 819
where i and j are the maps induces by i and π p above and δ is the map delivered by the
snake lemma
Now we can roll up those long exact sequences into an exact triangle
oplusH p +q (F p C lowast) oplusH p +q (F p C lowast)
oplusH p +q (F p C lowastF p minus1C lowast)
i
j
k
which gives us a spectral sequence E r pq
From now on we concentrate on spectral sequences arising from an exact couple which
complies the following two conditions we require D pq = 0 if p lt 0 and for q lt 0 we want
i D p minus1q +1 D pq to be an isomorphism
Remark ose two facts guarantee that the spectral sequence lives in the first quadrant
Proof First let p lt 0 Since k E pq D p minus1q = 0 has im(k ) = 0 we get d = jk = 0
Furthermore 0 = D pq j
E pq says that
E pq = ker(d )983087
im(d ) = ker(d ) = im( j ) = 0
Now we look at the case q lt 0 We have d E pq D p minus1q
j
D p minus1q and we know
that i D p minus1q sim D p q minus1 is an isomorphism But since im(k ) = ker(i ) = 0 this leads to
d = jk = 0 Moreover we have D p minus1q +1
i
sim D pq j
E pq and since ker( j ) = im(i ) =D p minus1q +1 we get
E pq = ker(d ) = ker(k ) = im( j ) = 0
Let Z r pq = ker(d r pq ) and B r pq = im(d r p +r q minusr +1) be the cycles and boundaries given by the
fact that the r -sheet of the spectral sequence contains chain complexes
Furthermore we set H n = limminusminusrarr
D p nminus p (the injective limit along the maps i D pq
D p +1q minus1 ie the disjoint union of the D p nminus p where two elements are said to be equal if
they are equal under compositions of i ) and F p H n = im(D pq H p +q ) by what we get
a filtration
F p minus1H n sube F p H n sube
since i (D p minus1nminus p minus1) sube D p nminus p is filtration is even canonically bounded D p nminus p = 0for p lt 0 causes F p H n = 0 If p gt n (which is equal to q = n minus p lt 0) we know that
D p n minus p D p +1nminus( p +1) and therefore H p +q = im(D p nminus p H p +q ) = F p H n
In particular the filtration satisfiescup
p F p H n = H n = limlarrminusminus
(H n F p H n) andcap
p F p H n = 0
So if a spectral sequence weakly converges to H lowast it even converges to H lowast
Proposition ere is a natural inclusion of F p H n F p minus1H n in E infin p nminus p e spectral se-
quence E r pq weakly converges to H lowast if and only if
Z infin =991785r
k minus1(i r D ) equals k minus1(0) = j (D )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 919
Proof First we define K pq to be the kernel of D pq H p +q Beeing an element of K pq is
equivalent to lying in the kernel of i r for an integer r so K pq =cup
r ker(i r ) It follows that
j (K pq ) =cup
r j (ker(i r )) =cup
r B r pq = B infin pq as one can easily see that the infinite boundaries
are the union of the B r pq (and similarly Z infin pq =cap
r Z r pq ) We look at the following diagram
0
0 K p minus1q +1 D p minusq q +1 F p minus1H n
0
0 K pq D pq F p H n 0
coker(i |K p minus1q +1)
coker(i ) F p H nF p minus1H n
i |K p minus1q +1 i
with short exact sequences as rows
Before applying the snake lemma we regard the cokernels
D pq 983087i (D p minus1q +1) = D pq
983087im(i ) = D pq
983087ker( j )
tells us that coker(i ) = j (D pq ) Furthermore we have im(i |K p minus1q +1) = ker( j |K pq ) () and
therefore coker(i |K p minus1q +1) j (K pq ) = B infin pq e snake lemma now says that
0 B infin pq j (D pq ) F p H n
983087F p minus1H n
0
is a short exact sequence
For all r isin Z we have
j (D pq ) = im( j ) = ker(k ) = k minus1(0) sube k minus1(i r D p minusr minus1q +r ) = Z r pq
what means j (D pq ) subecap
r Z r pq = Z infin pq We get a natural inclusion
F p H n983087F p minusq H n
j (D pq )1048623B infin pq sube Z infin pq
1048623B infin pq = E infin pq
and in particular we know now that the spectral sequences weakly converges to H lowast ie
F p H n F p minus1H n = E infin pq if and only if j (D ) = Z infin
Corollary (Convergence eorem) A spectral sequence with the requirements mentioned
above converges to H lowast = limminusminusrarr D E r pq rArr H p +q
If a spectral sequence arises from an exact couple of a complex C lowast with canonically bounded
filtration it converges to limminusminusrarr
D = H lowast which is the homology of the filtered complex C lowast
Proof A remark above tells us that the spectral sequence converges to H lowast if it weakly con-
verges to H lowast All we have to show is that Z infin = k minus1(0) if the spectral sequence complies
the two facts that D pq = 0 for p lt 0 and i D p minus1q +1
sim D pq for q lt 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1019
Let p and q be arbitrary integers ere exists an integer r such that p minus r minus 1 lt 0 which
implies i r (D p minusr minus1q +r ) = i r (0) = 0 We now see that Z r pq = k minus1(i r (D p minusr minus1q +r ) =
k minus1(0) en Z infin pq =cap
r prime Z r prime
pq = k minus1(0) and therefore Z infin = k minus1(0)
at means E r pq rArr H p +q
Let F p minus1C lowast sube F p C lowast sube sube C lowast be a canonically bounded filtration of a complex C lowastSince F p C n = 0 for all p lt 0 we get D pq = H p +q (F p C lowast) = 0 for those p For q lt 0 we have
p = n minusq gt n and F p C n = C n is leads to D pq = H n(F p C lowast) = H n(C lowast) = H n(F p +1C lowast) =D p +1q minus1 = D p +1q minus2 Alltogether it follows that a spectral sequence arising from a
canonically bounded complex converges to H lowast
It remains to show hat H lowast is the homology of C lowast
Let p lt n we then know that i is an isomorphism and therefore we get
H p +q im(D p nminus p H p +q ) D p nminus p = H p +q (F p C lowast) = H p +q (C lowast)
3 Double ComplexesIn Example we explained that a filtered complex yields an exact couple which in turn
yields a spectral sequence We will now go one step further and introduce double complexes
which yield filtered complexes and finally spectral sequences
Definition (Double complex) A double complex C lowastlowast is a family (C pq ) p q isinZ of objects
of an abelian category together with maps
d h pq C pq C p minus1q d v pq C pq C p q minus1
such that for each q resp p the families (C lowastq d hlowastq ) resp (C p lowast d v p lowast) are chain complexes
and additionally in every square the relation
d v p minus1q d h pq = plusmn d h p q minus1 d v pq
is satisfied In case of a ldquo+rdquo in this relation we call the double complex commutative and
in case of a ldquominusrdquo we call it anticommutative We say that a double complex lives in the first
quadrant if C pq = 0 whenever p lt 0 or q lt 0
We can picture a double complex like this (here we have a first quadrant double complex)
C 02
C 12
C 22
C 01 C 11 C 21
C 00 C 10 C 20
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1119
Definition Let C lowastlowast be an anticommutative double complex en we define the total
complex aached to C lowastlowast as the chain complex Totlowast(C lowastlowast) with entries⁴
Totn(C lowastlowast) =991893
p +q =n
C pq
and differentials d = d v + d h
A straightforward calculation using anticommutativity shows d 2 = 0 so that the total
complex is indeed a chain complex
We can turn the total complex of a given double complex into a filtered chain complex by
considering the following two obvious filtrations
ldquoFiltration by rowsrdquo
0 0 0
lowast lowast lowast
lowast lowast lowast
ldquoFiltration by columnsrdquo
lowast lowast 0lowast lowast 0lowast lowast 0
More precisely we set
F (R) p Totn(C lowastlowast) =
991893i le p
C i nminusi F (C)q Totn(C lowastlowast) =
991893 j leq
C nminus j j
In both cases we get
983080F p Totlowast(C lowastlowast)
983087F p minus1 Totlowast(C lowastlowast)
983081n
= C pq
for the n-th entry of the filtration quotients (where p + q = n) Note that when the double
complex lives in the first quadrant then both filtrations are canonically bounded ereforewe will assume from now on that every double complex lives in the first quadrant
e machinery developped in the previous sections now yields two spectral sequences
which we call (R)E r lowastlowast and (C)E r lowastlowast respectively By Corollary both spectral sequences
converge to the homology of the total complex Totlowast(C lowastlowast) Oen one is not really inter-
ested in this homology but one can play these sequences off against one another to obtain
information about the entries of the spectral sequence
Can we write down more explicitly how these spectral sequences look like We have to
apply the technique of exact couples to our situation Doing so we obtain
Proposition e E 1 page of (C)E r lowastlowast is obtained from the initial double complex C lowastlowast by
taking homology in vertical direction and the differentials d 1 on that page are the maps in-
duced by the horizontal differentials d h of C lowastlowast on this homology Dually the E 1 page of (R)E r lowastlowastis obtained by ldquotransposingrdquo this procedure
⁴ It is also common to consider a slightly differently defined total complex in which the direct sum is replacedby a direct product As we will mostly consider double complexes living in the first quadrant this does notmake a difference for us
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1219
Idea of proof To prove this we would have to go into the details of how the long homology
sequence coming from the short exact sequence of chain complexes
0 F p minus1 Totlowast(C lowastlowast) F p Totlowast(C lowastlowast) F p Totlowast(C lowastlowast)1048623F p minus1 Totlowast(C lowastlowast) 0
is constructed which in the end boils down to investigating the definition of the connectinghomomorphism coming from the snake lemma As this is a rather tedious calculation we
will omit this here
is proposition suggests to view the initial double complex C lowastlowast as the E 0 page of either
spectral sequence (possibly ldquotransposedrdquo)
Before giving some examples of applications of this construction we record the following
useful result
Lemma If we have a converging spectral sequence
E r pq rArr H p +q
and the E 2 page consists of 0 everywhere except in the boom row then we have
E 2 p 0 = H p
Proof is follows directly from Remark
Finally we note that we can also apply this whole technique to a commutative double com-
plex We can turn any such double complex into an anticommutative one by inserting some
minus1rsquos for example at every second horizontal or vertical arrow in this way in every square
exactly one map is multiplied by minus1 is does not affect the remaining properties of being
a double complex and does also not change homology
4 Examples
We conclude this note with a few examples that demonstrate the power of our techniques
e first examples are very simple whereas the later ones are a bit more interesting
41 The five lemma
As a first example we want to prove the five lemma Given have a commutative diagram
A B C D E
Aprime B prime C prime D prime E prime
f
д
f prime
д prime
α β γ δ ε
with exact rows we want to prove
(a) When β δ are monomorphisms and α is an epimorphism then γ is a monomorphism
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 419
Proof First we draw a picture of the 2-sheet of the spectral sequence
0
0
E 203
E 213
0
0
0
0
E 2
02
E 2
12
0
0
0 0 E 201 E 211 0 0
0
0 E 200 E 210
0
0
We see that E 2 pq = E infin pq and since we know that the spectral sequence converges we know
that there exists a filtration F p H n which fulfills E infin pq F p H p +q F p minus1H p +q
If p 0 1 we get 0 = E 2 pq = F p H p +q F p minusq H p +q which tells us F p H p +q = F p minus1H p +q and
therefore the filtration looks like
F minus2H n = F minus1H n sube F 0H n sube F 1H n = F 2H n =
Moreover we know thatcap
F p H n = 0 so F minus1H n = F minus2H n = = 0 and sincecup
F p H n =H n we get F 1H n = F 2H n = H n
Now let p = 0 We notice that
E 20n = E infin0n F 0H n
983087F minus1H n = F 0H n
For p = 1 we get
E 21nminus1 = E infin1nminus1 F 1H n 983087F 0H n = H n 983087F 0H n
and obviously the short exact sequence
0 F 0H n H n H n983087F 0H n
0
turns into the short exact sequence
0 E 20n H n E 21nminus1
0
Remark Let a spectral sequence converging to H lowast have E 2 pq = 0 unless q = 0 1 en
there is a long exact sequence
H p +1 E 2 p +10
d E 2 p minus11
H p E 2 p 0
d E 2 p minus21
H p minus1
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 519
Proof Again we first draw a picture
0
0
0
0
E 2minus11
E 2
01
E 2
11
E 2
21
E 2minus10 E 200 E 210 E 220
0
0
0
0
and we see that the objects at the 3-sheet are the objects E infin pq For q 1 0 we get
F p H p +q 983087F p minus1H p +q E
infin pq = E 3 pq = H (E 2 pq ) = ker(d 2 pq )
1048623im(d 2 p +2q minus1) = 0
and therefore F p H p +q = F p minus1H p +q
Letrsquos have a look at the filtration of H n
= F nminus3H n = F nminus2H n sube F nminus1H n sube F nH n = F n+1H n =
capF p H n = 0 implies F r H n = 0 for r le p +q minus 2 and
cupF p H n = H n implies F r H n = H n for
r ge n = p + q
Let q = 0 We get
H p 983087F p minus1H p = F p H p
983087F p minus1H p E
infin p 0 = E 3 p 0 = ker(d 2 p 0)
1048623im(d 2 p +2minus1) = ker(d 2 p 0)
And q = 1 leads to
F p H p +1 = F p H p +1983087F p minus1H p +1 E
infin p 1 = E 3 p 1
= H (E 2 p 1) = ker (d 2 p 1)1048623
im(d 2 p +20) = E 2 p 1
1048623im(d 2 p +20)
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 619
We can now put all this together to a long exact sequence
0
0
E 2 p +10 kerd 2 p +10
H p F p minus1H p
H p +1 E 2 p +10 E 2 p minus11 H p
kerd 2 p +10 F p minus1H p
0 0 0 0
2 Construction of Spectral Sequences
In this section we introduce the important notions of exact couples and filtered complexes
Every exact couple yields a spectral sequence and every filtered complex yields an exact
couple Finally we prove a convergence theorem for spectral sequences obtained in this
way
Construction (Exact couples and their derivations) An exact couple is a pair of objects
D and E of C and morphisms i j and k such that the diagram
D
D
E
i
j k
is exact at each vertex Since d = jk complies with d 2 = ( jk )2 = j (k j )k = j 0k = 0 we
can apply homology by seing H (E ) = ker(d )im(d ) and get the derived exact couple
i (D ) i (D )
H (E )
i |i (D )
j (2)k (2)
with j (2)(i (x )) = [ j (x )] and k (2)([e ]) = k (e ) x isin D and [e ] isin H (E ) It is an easy computa-
tion that j (2) and k (2) are well-defined and that the derived couple is exact It suggests itself
to iterate this process and set E 1 = E and E r = H (E r minus1) d 1 = d = jk and d r = j (r )k (r )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 719
e (r + 1)-th exact couple looks like
i r (D ) i r (D )
E r +1
i |i r (D )
j (r +1)k (r +1)
Remark Notice that the cycles can be described as Z r = k minus1(i r (D )) and the boundaries
as B r = j (ker(i r ))
Proof One just calculates
Z r = ker(d r ) = ker( j (r )k (r )) = ker( ji minusr +1k )
= k minus1(ker( ji minusr +1)) = k minus1(i r minus1(ker j ))
= k minus1(i r minus1(im i )) = k minus1(i r (D ))
B r = im(d r ) = im( j (r )k (r )) = im( ji minusr +1k )
= j (i minusr +1k (E pq )) = j (i minusr +1 im(k ))
= j (i minusr +1(ker(i )) = j (i minusr (0)) = j (ker(i r ))
We now regardD =oplus
D pq and E =oplus
E pq with the morphismsi j and k having bidegree
(1 minus1) (0 0)sup3 and (minus1 0) respectively (ie i (D pq ) sube D p +1q minus1 and so on) e bidegrees
of i and k do not change by passage to the derived couple while the bidegree of j (r ) in the
r -th couple is deg( j (r )) = deg( j ) minus (r minus 1) deg(i ) = (0 0) minus (r minus 1)(1 minus1) = (minus(r minus 1) r minus 1)since j (r )(i r minus1(x )) = [ j (x )] and therefore the bidegree of the differential d r = j (r )k (r ) is
deg(d r ) = (minusr r minus 1) just as claimed in the definition of spectral sequences
In this way the exact couple yields a spectral sequence with objects E
r
pq Example (Exact couple of a filtered complex) Let C lowast be a complex with a filtration
F p C lowast sube F p +1C lowast sube sube C lowast
such that there are integers s lt t for eachn with F s C n = 0 and F t C n = C n (Such a filtration
is called bounded Particularly this means F k C n = 0 for all k le s and F k C n = C n for all
k ge t We will always consider canonically bounded filtrations ie s = minus1 and F nC n = C nfor all n ndash what leads to a first quadrant spectral sequence)
We get short exact sequences
0 F p minus1C lowast
i
F p C lowast
π p
F p C lowast983087F p minus1C lowast
0
and by applying homology long exact secquences
H p minus1q +1(F p minus1C lowast)
i H p +q (F p C lowast)
j H p +q
983080F p C lowast
983087F p minus1C lowast
983081
δ H p minus1+q (F p C lowast)
sup3 To avoid confusion because of all the indices and variables we assume a = 0 ndash but j could also have bidegree(minusa a ) a 0 would not be more difficult but we would have to take care of it
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 819
where i and j are the maps induces by i and π p above and δ is the map delivered by the
snake lemma
Now we can roll up those long exact sequences into an exact triangle
oplusH p +q (F p C lowast) oplusH p +q (F p C lowast)
oplusH p +q (F p C lowastF p minus1C lowast)
i
j
k
which gives us a spectral sequence E r pq
From now on we concentrate on spectral sequences arising from an exact couple which
complies the following two conditions we require D pq = 0 if p lt 0 and for q lt 0 we want
i D p minus1q +1 D pq to be an isomorphism
Remark ose two facts guarantee that the spectral sequence lives in the first quadrant
Proof First let p lt 0 Since k E pq D p minus1q = 0 has im(k ) = 0 we get d = jk = 0
Furthermore 0 = D pq j
E pq says that
E pq = ker(d )983087
im(d ) = ker(d ) = im( j ) = 0
Now we look at the case q lt 0 We have d E pq D p minus1q
j
D p minus1q and we know
that i D p minus1q sim D p q minus1 is an isomorphism But since im(k ) = ker(i ) = 0 this leads to
d = jk = 0 Moreover we have D p minus1q +1
i
sim D pq j
E pq and since ker( j ) = im(i ) =D p minus1q +1 we get
E pq = ker(d ) = ker(k ) = im( j ) = 0
Let Z r pq = ker(d r pq ) and B r pq = im(d r p +r q minusr +1) be the cycles and boundaries given by the
fact that the r -sheet of the spectral sequence contains chain complexes
Furthermore we set H n = limminusminusrarr
D p nminus p (the injective limit along the maps i D pq
D p +1q minus1 ie the disjoint union of the D p nminus p where two elements are said to be equal if
they are equal under compositions of i ) and F p H n = im(D pq H p +q ) by what we get
a filtration
F p minus1H n sube F p H n sube
since i (D p minus1nminus p minus1) sube D p nminus p is filtration is even canonically bounded D p nminus p = 0for p lt 0 causes F p H n = 0 If p gt n (which is equal to q = n minus p lt 0) we know that
D p n minus p D p +1nminus( p +1) and therefore H p +q = im(D p nminus p H p +q ) = F p H n
In particular the filtration satisfiescup
p F p H n = H n = limlarrminusminus
(H n F p H n) andcap
p F p H n = 0
So if a spectral sequence weakly converges to H lowast it even converges to H lowast
Proposition ere is a natural inclusion of F p H n F p minus1H n in E infin p nminus p e spectral se-
quence E r pq weakly converges to H lowast if and only if
Z infin =991785r
k minus1(i r D ) equals k minus1(0) = j (D )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 919
Proof First we define K pq to be the kernel of D pq H p +q Beeing an element of K pq is
equivalent to lying in the kernel of i r for an integer r so K pq =cup
r ker(i r ) It follows that
j (K pq ) =cup
r j (ker(i r )) =cup
r B r pq = B infin pq as one can easily see that the infinite boundaries
are the union of the B r pq (and similarly Z infin pq =cap
r Z r pq ) We look at the following diagram
0
0 K p minus1q +1 D p minusq q +1 F p minus1H n
0
0 K pq D pq F p H n 0
coker(i |K p minus1q +1)
coker(i ) F p H nF p minus1H n
i |K p minus1q +1 i
with short exact sequences as rows
Before applying the snake lemma we regard the cokernels
D pq 983087i (D p minus1q +1) = D pq
983087im(i ) = D pq
983087ker( j )
tells us that coker(i ) = j (D pq ) Furthermore we have im(i |K p minus1q +1) = ker( j |K pq ) () and
therefore coker(i |K p minus1q +1) j (K pq ) = B infin pq e snake lemma now says that
0 B infin pq j (D pq ) F p H n
983087F p minus1H n
0
is a short exact sequence
For all r isin Z we have
j (D pq ) = im( j ) = ker(k ) = k minus1(0) sube k minus1(i r D p minusr minus1q +r ) = Z r pq
what means j (D pq ) subecap
r Z r pq = Z infin pq We get a natural inclusion
F p H n983087F p minusq H n
j (D pq )1048623B infin pq sube Z infin pq
1048623B infin pq = E infin pq
and in particular we know now that the spectral sequences weakly converges to H lowast ie
F p H n F p minus1H n = E infin pq if and only if j (D ) = Z infin
Corollary (Convergence eorem) A spectral sequence with the requirements mentioned
above converges to H lowast = limminusminusrarr D E r pq rArr H p +q
If a spectral sequence arises from an exact couple of a complex C lowast with canonically bounded
filtration it converges to limminusminusrarr
D = H lowast which is the homology of the filtered complex C lowast
Proof A remark above tells us that the spectral sequence converges to H lowast if it weakly con-
verges to H lowast All we have to show is that Z infin = k minus1(0) if the spectral sequence complies
the two facts that D pq = 0 for p lt 0 and i D p minus1q +1
sim D pq for q lt 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1019
Let p and q be arbitrary integers ere exists an integer r such that p minus r minus 1 lt 0 which
implies i r (D p minusr minus1q +r ) = i r (0) = 0 We now see that Z r pq = k minus1(i r (D p minusr minus1q +r ) =
k minus1(0) en Z infin pq =cap
r prime Z r prime
pq = k minus1(0) and therefore Z infin = k minus1(0)
at means E r pq rArr H p +q
Let F p minus1C lowast sube F p C lowast sube sube C lowast be a canonically bounded filtration of a complex C lowastSince F p C n = 0 for all p lt 0 we get D pq = H p +q (F p C lowast) = 0 for those p For q lt 0 we have
p = n minusq gt n and F p C n = C n is leads to D pq = H n(F p C lowast) = H n(C lowast) = H n(F p +1C lowast) =D p +1q minus1 = D p +1q minus2 Alltogether it follows that a spectral sequence arising from a
canonically bounded complex converges to H lowast
It remains to show hat H lowast is the homology of C lowast
Let p lt n we then know that i is an isomorphism and therefore we get
H p +q im(D p nminus p H p +q ) D p nminus p = H p +q (F p C lowast) = H p +q (C lowast)
3 Double ComplexesIn Example we explained that a filtered complex yields an exact couple which in turn
yields a spectral sequence We will now go one step further and introduce double complexes
which yield filtered complexes and finally spectral sequences
Definition (Double complex) A double complex C lowastlowast is a family (C pq ) p q isinZ of objects
of an abelian category together with maps
d h pq C pq C p minus1q d v pq C pq C p q minus1
such that for each q resp p the families (C lowastq d hlowastq ) resp (C p lowast d v p lowast) are chain complexes
and additionally in every square the relation
d v p minus1q d h pq = plusmn d h p q minus1 d v pq
is satisfied In case of a ldquo+rdquo in this relation we call the double complex commutative and
in case of a ldquominusrdquo we call it anticommutative We say that a double complex lives in the first
quadrant if C pq = 0 whenever p lt 0 or q lt 0
We can picture a double complex like this (here we have a first quadrant double complex)
C 02
C 12
C 22
C 01 C 11 C 21
C 00 C 10 C 20
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1119
Definition Let C lowastlowast be an anticommutative double complex en we define the total
complex aached to C lowastlowast as the chain complex Totlowast(C lowastlowast) with entries⁴
Totn(C lowastlowast) =991893
p +q =n
C pq
and differentials d = d v + d h
A straightforward calculation using anticommutativity shows d 2 = 0 so that the total
complex is indeed a chain complex
We can turn the total complex of a given double complex into a filtered chain complex by
considering the following two obvious filtrations
ldquoFiltration by rowsrdquo
0 0 0
lowast lowast lowast
lowast lowast lowast
ldquoFiltration by columnsrdquo
lowast lowast 0lowast lowast 0lowast lowast 0
More precisely we set
F (R) p Totn(C lowastlowast) =
991893i le p
C i nminusi F (C)q Totn(C lowastlowast) =
991893 j leq
C nminus j j
In both cases we get
983080F p Totlowast(C lowastlowast)
983087F p minus1 Totlowast(C lowastlowast)
983081n
= C pq
for the n-th entry of the filtration quotients (where p + q = n) Note that when the double
complex lives in the first quadrant then both filtrations are canonically bounded ereforewe will assume from now on that every double complex lives in the first quadrant
e machinery developped in the previous sections now yields two spectral sequences
which we call (R)E r lowastlowast and (C)E r lowastlowast respectively By Corollary both spectral sequences
converge to the homology of the total complex Totlowast(C lowastlowast) Oen one is not really inter-
ested in this homology but one can play these sequences off against one another to obtain
information about the entries of the spectral sequence
Can we write down more explicitly how these spectral sequences look like We have to
apply the technique of exact couples to our situation Doing so we obtain
Proposition e E 1 page of (C)E r lowastlowast is obtained from the initial double complex C lowastlowast by
taking homology in vertical direction and the differentials d 1 on that page are the maps in-
duced by the horizontal differentials d h of C lowastlowast on this homology Dually the E 1 page of (R)E r lowastlowastis obtained by ldquotransposingrdquo this procedure
⁴ It is also common to consider a slightly differently defined total complex in which the direct sum is replacedby a direct product As we will mostly consider double complexes living in the first quadrant this does notmake a difference for us
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1219
Idea of proof To prove this we would have to go into the details of how the long homology
sequence coming from the short exact sequence of chain complexes
0 F p minus1 Totlowast(C lowastlowast) F p Totlowast(C lowastlowast) F p Totlowast(C lowastlowast)1048623F p minus1 Totlowast(C lowastlowast) 0
is constructed which in the end boils down to investigating the definition of the connectinghomomorphism coming from the snake lemma As this is a rather tedious calculation we
will omit this here
is proposition suggests to view the initial double complex C lowastlowast as the E 0 page of either
spectral sequence (possibly ldquotransposedrdquo)
Before giving some examples of applications of this construction we record the following
useful result
Lemma If we have a converging spectral sequence
E r pq rArr H p +q
and the E 2 page consists of 0 everywhere except in the boom row then we have
E 2 p 0 = H p
Proof is follows directly from Remark
Finally we note that we can also apply this whole technique to a commutative double com-
plex We can turn any such double complex into an anticommutative one by inserting some
minus1rsquos for example at every second horizontal or vertical arrow in this way in every square
exactly one map is multiplied by minus1 is does not affect the remaining properties of being
a double complex and does also not change homology
4 Examples
We conclude this note with a few examples that demonstrate the power of our techniques
e first examples are very simple whereas the later ones are a bit more interesting
41 The five lemma
As a first example we want to prove the five lemma Given have a commutative diagram
A B C D E
Aprime B prime C prime D prime E prime
f
д
f prime
д prime
α β γ δ ε
with exact rows we want to prove
(a) When β δ are monomorphisms and α is an epimorphism then γ is a monomorphism
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 519
Proof Again we first draw a picture
0
0
0
0
E 2minus11
E 2
01
E 2
11
E 2
21
E 2minus10 E 200 E 210 E 220
0
0
0
0
and we see that the objects at the 3-sheet are the objects E infin pq For q 1 0 we get
F p H p +q 983087F p minus1H p +q E
infin pq = E 3 pq = H (E 2 pq ) = ker(d 2 pq )
1048623im(d 2 p +2q minus1) = 0
and therefore F p H p +q = F p minus1H p +q
Letrsquos have a look at the filtration of H n
= F nminus3H n = F nminus2H n sube F nminus1H n sube F nH n = F n+1H n =
capF p H n = 0 implies F r H n = 0 for r le p +q minus 2 and
cupF p H n = H n implies F r H n = H n for
r ge n = p + q
Let q = 0 We get
H p 983087F p minus1H p = F p H p
983087F p minus1H p E
infin p 0 = E 3 p 0 = ker(d 2 p 0)
1048623im(d 2 p +2minus1) = ker(d 2 p 0)
And q = 1 leads to
F p H p +1 = F p H p +1983087F p minus1H p +1 E
infin p 1 = E 3 p 1
= H (E 2 p 1) = ker (d 2 p 1)1048623
im(d 2 p +20) = E 2 p 1
1048623im(d 2 p +20)
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 619
We can now put all this together to a long exact sequence
0
0
E 2 p +10 kerd 2 p +10
H p F p minus1H p
H p +1 E 2 p +10 E 2 p minus11 H p
kerd 2 p +10 F p minus1H p
0 0 0 0
2 Construction of Spectral Sequences
In this section we introduce the important notions of exact couples and filtered complexes
Every exact couple yields a spectral sequence and every filtered complex yields an exact
couple Finally we prove a convergence theorem for spectral sequences obtained in this
way
Construction (Exact couples and their derivations) An exact couple is a pair of objects
D and E of C and morphisms i j and k such that the diagram
D
D
E
i
j k
is exact at each vertex Since d = jk complies with d 2 = ( jk )2 = j (k j )k = j 0k = 0 we
can apply homology by seing H (E ) = ker(d )im(d ) and get the derived exact couple
i (D ) i (D )
H (E )
i |i (D )
j (2)k (2)
with j (2)(i (x )) = [ j (x )] and k (2)([e ]) = k (e ) x isin D and [e ] isin H (E ) It is an easy computa-
tion that j (2) and k (2) are well-defined and that the derived couple is exact It suggests itself
to iterate this process and set E 1 = E and E r = H (E r minus1) d 1 = d = jk and d r = j (r )k (r )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 719
e (r + 1)-th exact couple looks like
i r (D ) i r (D )
E r +1
i |i r (D )
j (r +1)k (r +1)
Remark Notice that the cycles can be described as Z r = k minus1(i r (D )) and the boundaries
as B r = j (ker(i r ))
Proof One just calculates
Z r = ker(d r ) = ker( j (r )k (r )) = ker( ji minusr +1k )
= k minus1(ker( ji minusr +1)) = k minus1(i r minus1(ker j ))
= k minus1(i r minus1(im i )) = k minus1(i r (D ))
B r = im(d r ) = im( j (r )k (r )) = im( ji minusr +1k )
= j (i minusr +1k (E pq )) = j (i minusr +1 im(k ))
= j (i minusr +1(ker(i )) = j (i minusr (0)) = j (ker(i r ))
We now regardD =oplus
D pq and E =oplus
E pq with the morphismsi j and k having bidegree
(1 minus1) (0 0)sup3 and (minus1 0) respectively (ie i (D pq ) sube D p +1q minus1 and so on) e bidegrees
of i and k do not change by passage to the derived couple while the bidegree of j (r ) in the
r -th couple is deg( j (r )) = deg( j ) minus (r minus 1) deg(i ) = (0 0) minus (r minus 1)(1 minus1) = (minus(r minus 1) r minus 1)since j (r )(i r minus1(x )) = [ j (x )] and therefore the bidegree of the differential d r = j (r )k (r ) is
deg(d r ) = (minusr r minus 1) just as claimed in the definition of spectral sequences
In this way the exact couple yields a spectral sequence with objects E
r
pq Example (Exact couple of a filtered complex) Let C lowast be a complex with a filtration
F p C lowast sube F p +1C lowast sube sube C lowast
such that there are integers s lt t for eachn with F s C n = 0 and F t C n = C n (Such a filtration
is called bounded Particularly this means F k C n = 0 for all k le s and F k C n = C n for all
k ge t We will always consider canonically bounded filtrations ie s = minus1 and F nC n = C nfor all n ndash what leads to a first quadrant spectral sequence)
We get short exact sequences
0 F p minus1C lowast
i
F p C lowast
π p
F p C lowast983087F p minus1C lowast
0
and by applying homology long exact secquences
H p minus1q +1(F p minus1C lowast)
i H p +q (F p C lowast)
j H p +q
983080F p C lowast
983087F p minus1C lowast
983081
δ H p minus1+q (F p C lowast)
sup3 To avoid confusion because of all the indices and variables we assume a = 0 ndash but j could also have bidegree(minusa a ) a 0 would not be more difficult but we would have to take care of it
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 819
where i and j are the maps induces by i and π p above and δ is the map delivered by the
snake lemma
Now we can roll up those long exact sequences into an exact triangle
oplusH p +q (F p C lowast) oplusH p +q (F p C lowast)
oplusH p +q (F p C lowastF p minus1C lowast)
i
j
k
which gives us a spectral sequence E r pq
From now on we concentrate on spectral sequences arising from an exact couple which
complies the following two conditions we require D pq = 0 if p lt 0 and for q lt 0 we want
i D p minus1q +1 D pq to be an isomorphism
Remark ose two facts guarantee that the spectral sequence lives in the first quadrant
Proof First let p lt 0 Since k E pq D p minus1q = 0 has im(k ) = 0 we get d = jk = 0
Furthermore 0 = D pq j
E pq says that
E pq = ker(d )983087
im(d ) = ker(d ) = im( j ) = 0
Now we look at the case q lt 0 We have d E pq D p minus1q
j
D p minus1q and we know
that i D p minus1q sim D p q minus1 is an isomorphism But since im(k ) = ker(i ) = 0 this leads to
d = jk = 0 Moreover we have D p minus1q +1
i
sim D pq j
E pq and since ker( j ) = im(i ) =D p minus1q +1 we get
E pq = ker(d ) = ker(k ) = im( j ) = 0
Let Z r pq = ker(d r pq ) and B r pq = im(d r p +r q minusr +1) be the cycles and boundaries given by the
fact that the r -sheet of the spectral sequence contains chain complexes
Furthermore we set H n = limminusminusrarr
D p nminus p (the injective limit along the maps i D pq
D p +1q minus1 ie the disjoint union of the D p nminus p where two elements are said to be equal if
they are equal under compositions of i ) and F p H n = im(D pq H p +q ) by what we get
a filtration
F p minus1H n sube F p H n sube
since i (D p minus1nminus p minus1) sube D p nminus p is filtration is even canonically bounded D p nminus p = 0for p lt 0 causes F p H n = 0 If p gt n (which is equal to q = n minus p lt 0) we know that
D p n minus p D p +1nminus( p +1) and therefore H p +q = im(D p nminus p H p +q ) = F p H n
In particular the filtration satisfiescup
p F p H n = H n = limlarrminusminus
(H n F p H n) andcap
p F p H n = 0
So if a spectral sequence weakly converges to H lowast it even converges to H lowast
Proposition ere is a natural inclusion of F p H n F p minus1H n in E infin p nminus p e spectral se-
quence E r pq weakly converges to H lowast if and only if
Z infin =991785r
k minus1(i r D ) equals k minus1(0) = j (D )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 919
Proof First we define K pq to be the kernel of D pq H p +q Beeing an element of K pq is
equivalent to lying in the kernel of i r for an integer r so K pq =cup
r ker(i r ) It follows that
j (K pq ) =cup
r j (ker(i r )) =cup
r B r pq = B infin pq as one can easily see that the infinite boundaries
are the union of the B r pq (and similarly Z infin pq =cap
r Z r pq ) We look at the following diagram
0
0 K p minus1q +1 D p minusq q +1 F p minus1H n
0
0 K pq D pq F p H n 0
coker(i |K p minus1q +1)
coker(i ) F p H nF p minus1H n
i |K p minus1q +1 i
with short exact sequences as rows
Before applying the snake lemma we regard the cokernels
D pq 983087i (D p minus1q +1) = D pq
983087im(i ) = D pq
983087ker( j )
tells us that coker(i ) = j (D pq ) Furthermore we have im(i |K p minus1q +1) = ker( j |K pq ) () and
therefore coker(i |K p minus1q +1) j (K pq ) = B infin pq e snake lemma now says that
0 B infin pq j (D pq ) F p H n
983087F p minus1H n
0
is a short exact sequence
For all r isin Z we have
j (D pq ) = im( j ) = ker(k ) = k minus1(0) sube k minus1(i r D p minusr minus1q +r ) = Z r pq
what means j (D pq ) subecap
r Z r pq = Z infin pq We get a natural inclusion
F p H n983087F p minusq H n
j (D pq )1048623B infin pq sube Z infin pq
1048623B infin pq = E infin pq
and in particular we know now that the spectral sequences weakly converges to H lowast ie
F p H n F p minus1H n = E infin pq if and only if j (D ) = Z infin
Corollary (Convergence eorem) A spectral sequence with the requirements mentioned
above converges to H lowast = limminusminusrarr D E r pq rArr H p +q
If a spectral sequence arises from an exact couple of a complex C lowast with canonically bounded
filtration it converges to limminusminusrarr
D = H lowast which is the homology of the filtered complex C lowast
Proof A remark above tells us that the spectral sequence converges to H lowast if it weakly con-
verges to H lowast All we have to show is that Z infin = k minus1(0) if the spectral sequence complies
the two facts that D pq = 0 for p lt 0 and i D p minus1q +1
sim D pq for q lt 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1019
Let p and q be arbitrary integers ere exists an integer r such that p minus r minus 1 lt 0 which
implies i r (D p minusr minus1q +r ) = i r (0) = 0 We now see that Z r pq = k minus1(i r (D p minusr minus1q +r ) =
k minus1(0) en Z infin pq =cap
r prime Z r prime
pq = k minus1(0) and therefore Z infin = k minus1(0)
at means E r pq rArr H p +q
Let F p minus1C lowast sube F p C lowast sube sube C lowast be a canonically bounded filtration of a complex C lowastSince F p C n = 0 for all p lt 0 we get D pq = H p +q (F p C lowast) = 0 for those p For q lt 0 we have
p = n minusq gt n and F p C n = C n is leads to D pq = H n(F p C lowast) = H n(C lowast) = H n(F p +1C lowast) =D p +1q minus1 = D p +1q minus2 Alltogether it follows that a spectral sequence arising from a
canonically bounded complex converges to H lowast
It remains to show hat H lowast is the homology of C lowast
Let p lt n we then know that i is an isomorphism and therefore we get
H p +q im(D p nminus p H p +q ) D p nminus p = H p +q (F p C lowast) = H p +q (C lowast)
3 Double ComplexesIn Example we explained that a filtered complex yields an exact couple which in turn
yields a spectral sequence We will now go one step further and introduce double complexes
which yield filtered complexes and finally spectral sequences
Definition (Double complex) A double complex C lowastlowast is a family (C pq ) p q isinZ of objects
of an abelian category together with maps
d h pq C pq C p minus1q d v pq C pq C p q minus1
such that for each q resp p the families (C lowastq d hlowastq ) resp (C p lowast d v p lowast) are chain complexes
and additionally in every square the relation
d v p minus1q d h pq = plusmn d h p q minus1 d v pq
is satisfied In case of a ldquo+rdquo in this relation we call the double complex commutative and
in case of a ldquominusrdquo we call it anticommutative We say that a double complex lives in the first
quadrant if C pq = 0 whenever p lt 0 or q lt 0
We can picture a double complex like this (here we have a first quadrant double complex)
C 02
C 12
C 22
C 01 C 11 C 21
C 00 C 10 C 20
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1119
Definition Let C lowastlowast be an anticommutative double complex en we define the total
complex aached to C lowastlowast as the chain complex Totlowast(C lowastlowast) with entries⁴
Totn(C lowastlowast) =991893
p +q =n
C pq
and differentials d = d v + d h
A straightforward calculation using anticommutativity shows d 2 = 0 so that the total
complex is indeed a chain complex
We can turn the total complex of a given double complex into a filtered chain complex by
considering the following two obvious filtrations
ldquoFiltration by rowsrdquo
0 0 0
lowast lowast lowast
lowast lowast lowast
ldquoFiltration by columnsrdquo
lowast lowast 0lowast lowast 0lowast lowast 0
More precisely we set
F (R) p Totn(C lowastlowast) =
991893i le p
C i nminusi F (C)q Totn(C lowastlowast) =
991893 j leq
C nminus j j
In both cases we get
983080F p Totlowast(C lowastlowast)
983087F p minus1 Totlowast(C lowastlowast)
983081n
= C pq
for the n-th entry of the filtration quotients (where p + q = n) Note that when the double
complex lives in the first quadrant then both filtrations are canonically bounded ereforewe will assume from now on that every double complex lives in the first quadrant
e machinery developped in the previous sections now yields two spectral sequences
which we call (R)E r lowastlowast and (C)E r lowastlowast respectively By Corollary both spectral sequences
converge to the homology of the total complex Totlowast(C lowastlowast) Oen one is not really inter-
ested in this homology but one can play these sequences off against one another to obtain
information about the entries of the spectral sequence
Can we write down more explicitly how these spectral sequences look like We have to
apply the technique of exact couples to our situation Doing so we obtain
Proposition e E 1 page of (C)E r lowastlowast is obtained from the initial double complex C lowastlowast by
taking homology in vertical direction and the differentials d 1 on that page are the maps in-
duced by the horizontal differentials d h of C lowastlowast on this homology Dually the E 1 page of (R)E r lowastlowastis obtained by ldquotransposingrdquo this procedure
⁴ It is also common to consider a slightly differently defined total complex in which the direct sum is replacedby a direct product As we will mostly consider double complexes living in the first quadrant this does notmake a difference for us
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1219
Idea of proof To prove this we would have to go into the details of how the long homology
sequence coming from the short exact sequence of chain complexes
0 F p minus1 Totlowast(C lowastlowast) F p Totlowast(C lowastlowast) F p Totlowast(C lowastlowast)1048623F p minus1 Totlowast(C lowastlowast) 0
is constructed which in the end boils down to investigating the definition of the connectinghomomorphism coming from the snake lemma As this is a rather tedious calculation we
will omit this here
is proposition suggests to view the initial double complex C lowastlowast as the E 0 page of either
spectral sequence (possibly ldquotransposedrdquo)
Before giving some examples of applications of this construction we record the following
useful result
Lemma If we have a converging spectral sequence
E r pq rArr H p +q
and the E 2 page consists of 0 everywhere except in the boom row then we have
E 2 p 0 = H p
Proof is follows directly from Remark
Finally we note that we can also apply this whole technique to a commutative double com-
plex We can turn any such double complex into an anticommutative one by inserting some
minus1rsquos for example at every second horizontal or vertical arrow in this way in every square
exactly one map is multiplied by minus1 is does not affect the remaining properties of being
a double complex and does also not change homology
4 Examples
We conclude this note with a few examples that demonstrate the power of our techniques
e first examples are very simple whereas the later ones are a bit more interesting
41 The five lemma
As a first example we want to prove the five lemma Given have a commutative diagram
A B C D E
Aprime B prime C prime D prime E prime
f
д
f prime
д prime
α β γ δ ε
with exact rows we want to prove
(a) When β δ are monomorphisms and α is an epimorphism then γ is a monomorphism
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 619
We can now put all this together to a long exact sequence
0
0
E 2 p +10 kerd 2 p +10
H p F p minus1H p
H p +1 E 2 p +10 E 2 p minus11 H p
kerd 2 p +10 F p minus1H p
0 0 0 0
2 Construction of Spectral Sequences
In this section we introduce the important notions of exact couples and filtered complexes
Every exact couple yields a spectral sequence and every filtered complex yields an exact
couple Finally we prove a convergence theorem for spectral sequences obtained in this
way
Construction (Exact couples and their derivations) An exact couple is a pair of objects
D and E of C and morphisms i j and k such that the diagram
D
D
E
i
j k
is exact at each vertex Since d = jk complies with d 2 = ( jk )2 = j (k j )k = j 0k = 0 we
can apply homology by seing H (E ) = ker(d )im(d ) and get the derived exact couple
i (D ) i (D )
H (E )
i |i (D )
j (2)k (2)
with j (2)(i (x )) = [ j (x )] and k (2)([e ]) = k (e ) x isin D and [e ] isin H (E ) It is an easy computa-
tion that j (2) and k (2) are well-defined and that the derived couple is exact It suggests itself
to iterate this process and set E 1 = E and E r = H (E r minus1) d 1 = d = jk and d r = j (r )k (r )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 719
e (r + 1)-th exact couple looks like
i r (D ) i r (D )
E r +1
i |i r (D )
j (r +1)k (r +1)
Remark Notice that the cycles can be described as Z r = k minus1(i r (D )) and the boundaries
as B r = j (ker(i r ))
Proof One just calculates
Z r = ker(d r ) = ker( j (r )k (r )) = ker( ji minusr +1k )
= k minus1(ker( ji minusr +1)) = k minus1(i r minus1(ker j ))
= k minus1(i r minus1(im i )) = k minus1(i r (D ))
B r = im(d r ) = im( j (r )k (r )) = im( ji minusr +1k )
= j (i minusr +1k (E pq )) = j (i minusr +1 im(k ))
= j (i minusr +1(ker(i )) = j (i minusr (0)) = j (ker(i r ))
We now regardD =oplus
D pq and E =oplus
E pq with the morphismsi j and k having bidegree
(1 minus1) (0 0)sup3 and (minus1 0) respectively (ie i (D pq ) sube D p +1q minus1 and so on) e bidegrees
of i and k do not change by passage to the derived couple while the bidegree of j (r ) in the
r -th couple is deg( j (r )) = deg( j ) minus (r minus 1) deg(i ) = (0 0) minus (r minus 1)(1 minus1) = (minus(r minus 1) r minus 1)since j (r )(i r minus1(x )) = [ j (x )] and therefore the bidegree of the differential d r = j (r )k (r ) is
deg(d r ) = (minusr r minus 1) just as claimed in the definition of spectral sequences
In this way the exact couple yields a spectral sequence with objects E
r
pq Example (Exact couple of a filtered complex) Let C lowast be a complex with a filtration
F p C lowast sube F p +1C lowast sube sube C lowast
such that there are integers s lt t for eachn with F s C n = 0 and F t C n = C n (Such a filtration
is called bounded Particularly this means F k C n = 0 for all k le s and F k C n = C n for all
k ge t We will always consider canonically bounded filtrations ie s = minus1 and F nC n = C nfor all n ndash what leads to a first quadrant spectral sequence)
We get short exact sequences
0 F p minus1C lowast
i
F p C lowast
π p
F p C lowast983087F p minus1C lowast
0
and by applying homology long exact secquences
H p minus1q +1(F p minus1C lowast)
i H p +q (F p C lowast)
j H p +q
983080F p C lowast
983087F p minus1C lowast
983081
δ H p minus1+q (F p C lowast)
sup3 To avoid confusion because of all the indices and variables we assume a = 0 ndash but j could also have bidegree(minusa a ) a 0 would not be more difficult but we would have to take care of it
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 819
where i and j are the maps induces by i and π p above and δ is the map delivered by the
snake lemma
Now we can roll up those long exact sequences into an exact triangle
oplusH p +q (F p C lowast) oplusH p +q (F p C lowast)
oplusH p +q (F p C lowastF p minus1C lowast)
i
j
k
which gives us a spectral sequence E r pq
From now on we concentrate on spectral sequences arising from an exact couple which
complies the following two conditions we require D pq = 0 if p lt 0 and for q lt 0 we want
i D p minus1q +1 D pq to be an isomorphism
Remark ose two facts guarantee that the spectral sequence lives in the first quadrant
Proof First let p lt 0 Since k E pq D p minus1q = 0 has im(k ) = 0 we get d = jk = 0
Furthermore 0 = D pq j
E pq says that
E pq = ker(d )983087
im(d ) = ker(d ) = im( j ) = 0
Now we look at the case q lt 0 We have d E pq D p minus1q
j
D p minus1q and we know
that i D p minus1q sim D p q minus1 is an isomorphism But since im(k ) = ker(i ) = 0 this leads to
d = jk = 0 Moreover we have D p minus1q +1
i
sim D pq j
E pq and since ker( j ) = im(i ) =D p minus1q +1 we get
E pq = ker(d ) = ker(k ) = im( j ) = 0
Let Z r pq = ker(d r pq ) and B r pq = im(d r p +r q minusr +1) be the cycles and boundaries given by the
fact that the r -sheet of the spectral sequence contains chain complexes
Furthermore we set H n = limminusminusrarr
D p nminus p (the injective limit along the maps i D pq
D p +1q minus1 ie the disjoint union of the D p nminus p where two elements are said to be equal if
they are equal under compositions of i ) and F p H n = im(D pq H p +q ) by what we get
a filtration
F p minus1H n sube F p H n sube
since i (D p minus1nminus p minus1) sube D p nminus p is filtration is even canonically bounded D p nminus p = 0for p lt 0 causes F p H n = 0 If p gt n (which is equal to q = n minus p lt 0) we know that
D p n minus p D p +1nminus( p +1) and therefore H p +q = im(D p nminus p H p +q ) = F p H n
In particular the filtration satisfiescup
p F p H n = H n = limlarrminusminus
(H n F p H n) andcap
p F p H n = 0
So if a spectral sequence weakly converges to H lowast it even converges to H lowast
Proposition ere is a natural inclusion of F p H n F p minus1H n in E infin p nminus p e spectral se-
quence E r pq weakly converges to H lowast if and only if
Z infin =991785r
k minus1(i r D ) equals k minus1(0) = j (D )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 919
Proof First we define K pq to be the kernel of D pq H p +q Beeing an element of K pq is
equivalent to lying in the kernel of i r for an integer r so K pq =cup
r ker(i r ) It follows that
j (K pq ) =cup
r j (ker(i r )) =cup
r B r pq = B infin pq as one can easily see that the infinite boundaries
are the union of the B r pq (and similarly Z infin pq =cap
r Z r pq ) We look at the following diagram
0
0 K p minus1q +1 D p minusq q +1 F p minus1H n
0
0 K pq D pq F p H n 0
coker(i |K p minus1q +1)
coker(i ) F p H nF p minus1H n
i |K p minus1q +1 i
with short exact sequences as rows
Before applying the snake lemma we regard the cokernels
D pq 983087i (D p minus1q +1) = D pq
983087im(i ) = D pq
983087ker( j )
tells us that coker(i ) = j (D pq ) Furthermore we have im(i |K p minus1q +1) = ker( j |K pq ) () and
therefore coker(i |K p minus1q +1) j (K pq ) = B infin pq e snake lemma now says that
0 B infin pq j (D pq ) F p H n
983087F p minus1H n
0
is a short exact sequence
For all r isin Z we have
j (D pq ) = im( j ) = ker(k ) = k minus1(0) sube k minus1(i r D p minusr minus1q +r ) = Z r pq
what means j (D pq ) subecap
r Z r pq = Z infin pq We get a natural inclusion
F p H n983087F p minusq H n
j (D pq )1048623B infin pq sube Z infin pq
1048623B infin pq = E infin pq
and in particular we know now that the spectral sequences weakly converges to H lowast ie
F p H n F p minus1H n = E infin pq if and only if j (D ) = Z infin
Corollary (Convergence eorem) A spectral sequence with the requirements mentioned
above converges to H lowast = limminusminusrarr D E r pq rArr H p +q
If a spectral sequence arises from an exact couple of a complex C lowast with canonically bounded
filtration it converges to limminusminusrarr
D = H lowast which is the homology of the filtered complex C lowast
Proof A remark above tells us that the spectral sequence converges to H lowast if it weakly con-
verges to H lowast All we have to show is that Z infin = k minus1(0) if the spectral sequence complies
the two facts that D pq = 0 for p lt 0 and i D p minus1q +1
sim D pq for q lt 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1019
Let p and q be arbitrary integers ere exists an integer r such that p minus r minus 1 lt 0 which
implies i r (D p minusr minus1q +r ) = i r (0) = 0 We now see that Z r pq = k minus1(i r (D p minusr minus1q +r ) =
k minus1(0) en Z infin pq =cap
r prime Z r prime
pq = k minus1(0) and therefore Z infin = k minus1(0)
at means E r pq rArr H p +q
Let F p minus1C lowast sube F p C lowast sube sube C lowast be a canonically bounded filtration of a complex C lowastSince F p C n = 0 for all p lt 0 we get D pq = H p +q (F p C lowast) = 0 for those p For q lt 0 we have
p = n minusq gt n and F p C n = C n is leads to D pq = H n(F p C lowast) = H n(C lowast) = H n(F p +1C lowast) =D p +1q minus1 = D p +1q minus2 Alltogether it follows that a spectral sequence arising from a
canonically bounded complex converges to H lowast
It remains to show hat H lowast is the homology of C lowast
Let p lt n we then know that i is an isomorphism and therefore we get
H p +q im(D p nminus p H p +q ) D p nminus p = H p +q (F p C lowast) = H p +q (C lowast)
3 Double ComplexesIn Example we explained that a filtered complex yields an exact couple which in turn
yields a spectral sequence We will now go one step further and introduce double complexes
which yield filtered complexes and finally spectral sequences
Definition (Double complex) A double complex C lowastlowast is a family (C pq ) p q isinZ of objects
of an abelian category together with maps
d h pq C pq C p minus1q d v pq C pq C p q minus1
such that for each q resp p the families (C lowastq d hlowastq ) resp (C p lowast d v p lowast) are chain complexes
and additionally in every square the relation
d v p minus1q d h pq = plusmn d h p q minus1 d v pq
is satisfied In case of a ldquo+rdquo in this relation we call the double complex commutative and
in case of a ldquominusrdquo we call it anticommutative We say that a double complex lives in the first
quadrant if C pq = 0 whenever p lt 0 or q lt 0
We can picture a double complex like this (here we have a first quadrant double complex)
C 02
C 12
C 22
C 01 C 11 C 21
C 00 C 10 C 20
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1119
Definition Let C lowastlowast be an anticommutative double complex en we define the total
complex aached to C lowastlowast as the chain complex Totlowast(C lowastlowast) with entries⁴
Totn(C lowastlowast) =991893
p +q =n
C pq
and differentials d = d v + d h
A straightforward calculation using anticommutativity shows d 2 = 0 so that the total
complex is indeed a chain complex
We can turn the total complex of a given double complex into a filtered chain complex by
considering the following two obvious filtrations
ldquoFiltration by rowsrdquo
0 0 0
lowast lowast lowast
lowast lowast lowast
ldquoFiltration by columnsrdquo
lowast lowast 0lowast lowast 0lowast lowast 0
More precisely we set
F (R) p Totn(C lowastlowast) =
991893i le p
C i nminusi F (C)q Totn(C lowastlowast) =
991893 j leq
C nminus j j
In both cases we get
983080F p Totlowast(C lowastlowast)
983087F p minus1 Totlowast(C lowastlowast)
983081n
= C pq
for the n-th entry of the filtration quotients (where p + q = n) Note that when the double
complex lives in the first quadrant then both filtrations are canonically bounded ereforewe will assume from now on that every double complex lives in the first quadrant
e machinery developped in the previous sections now yields two spectral sequences
which we call (R)E r lowastlowast and (C)E r lowastlowast respectively By Corollary both spectral sequences
converge to the homology of the total complex Totlowast(C lowastlowast) Oen one is not really inter-
ested in this homology but one can play these sequences off against one another to obtain
information about the entries of the spectral sequence
Can we write down more explicitly how these spectral sequences look like We have to
apply the technique of exact couples to our situation Doing so we obtain
Proposition e E 1 page of (C)E r lowastlowast is obtained from the initial double complex C lowastlowast by
taking homology in vertical direction and the differentials d 1 on that page are the maps in-
duced by the horizontal differentials d h of C lowastlowast on this homology Dually the E 1 page of (R)E r lowastlowastis obtained by ldquotransposingrdquo this procedure
⁴ It is also common to consider a slightly differently defined total complex in which the direct sum is replacedby a direct product As we will mostly consider double complexes living in the first quadrant this does notmake a difference for us
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1219
Idea of proof To prove this we would have to go into the details of how the long homology
sequence coming from the short exact sequence of chain complexes
0 F p minus1 Totlowast(C lowastlowast) F p Totlowast(C lowastlowast) F p Totlowast(C lowastlowast)1048623F p minus1 Totlowast(C lowastlowast) 0
is constructed which in the end boils down to investigating the definition of the connectinghomomorphism coming from the snake lemma As this is a rather tedious calculation we
will omit this here
is proposition suggests to view the initial double complex C lowastlowast as the E 0 page of either
spectral sequence (possibly ldquotransposedrdquo)
Before giving some examples of applications of this construction we record the following
useful result
Lemma If we have a converging spectral sequence
E r pq rArr H p +q
and the E 2 page consists of 0 everywhere except in the boom row then we have
E 2 p 0 = H p
Proof is follows directly from Remark
Finally we note that we can also apply this whole technique to a commutative double com-
plex We can turn any such double complex into an anticommutative one by inserting some
minus1rsquos for example at every second horizontal or vertical arrow in this way in every square
exactly one map is multiplied by minus1 is does not affect the remaining properties of being
a double complex and does also not change homology
4 Examples
We conclude this note with a few examples that demonstrate the power of our techniques
e first examples are very simple whereas the later ones are a bit more interesting
41 The five lemma
As a first example we want to prove the five lemma Given have a commutative diagram
A B C D E
Aprime B prime C prime D prime E prime
f
д
f prime
д prime
α β γ δ ε
with exact rows we want to prove
(a) When β δ are monomorphisms and α is an epimorphism then γ is a monomorphism
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 719
e (r + 1)-th exact couple looks like
i r (D ) i r (D )
E r +1
i |i r (D )
j (r +1)k (r +1)
Remark Notice that the cycles can be described as Z r = k minus1(i r (D )) and the boundaries
as B r = j (ker(i r ))
Proof One just calculates
Z r = ker(d r ) = ker( j (r )k (r )) = ker( ji minusr +1k )
= k minus1(ker( ji minusr +1)) = k minus1(i r minus1(ker j ))
= k minus1(i r minus1(im i )) = k minus1(i r (D ))
B r = im(d r ) = im( j (r )k (r )) = im( ji minusr +1k )
= j (i minusr +1k (E pq )) = j (i minusr +1 im(k ))
= j (i minusr +1(ker(i )) = j (i minusr (0)) = j (ker(i r ))
We now regardD =oplus
D pq and E =oplus
E pq with the morphismsi j and k having bidegree
(1 minus1) (0 0)sup3 and (minus1 0) respectively (ie i (D pq ) sube D p +1q minus1 and so on) e bidegrees
of i and k do not change by passage to the derived couple while the bidegree of j (r ) in the
r -th couple is deg( j (r )) = deg( j ) minus (r minus 1) deg(i ) = (0 0) minus (r minus 1)(1 minus1) = (minus(r minus 1) r minus 1)since j (r )(i r minus1(x )) = [ j (x )] and therefore the bidegree of the differential d r = j (r )k (r ) is
deg(d r ) = (minusr r minus 1) just as claimed in the definition of spectral sequences
In this way the exact couple yields a spectral sequence with objects E
r
pq Example (Exact couple of a filtered complex) Let C lowast be a complex with a filtration
F p C lowast sube F p +1C lowast sube sube C lowast
such that there are integers s lt t for eachn with F s C n = 0 and F t C n = C n (Such a filtration
is called bounded Particularly this means F k C n = 0 for all k le s and F k C n = C n for all
k ge t We will always consider canonically bounded filtrations ie s = minus1 and F nC n = C nfor all n ndash what leads to a first quadrant spectral sequence)
We get short exact sequences
0 F p minus1C lowast
i
F p C lowast
π p
F p C lowast983087F p minus1C lowast
0
and by applying homology long exact secquences
H p minus1q +1(F p minus1C lowast)
i H p +q (F p C lowast)
j H p +q
983080F p C lowast
983087F p minus1C lowast
983081
δ H p minus1+q (F p C lowast)
sup3 To avoid confusion because of all the indices and variables we assume a = 0 ndash but j could also have bidegree(minusa a ) a 0 would not be more difficult but we would have to take care of it
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 819
where i and j are the maps induces by i and π p above and δ is the map delivered by the
snake lemma
Now we can roll up those long exact sequences into an exact triangle
oplusH p +q (F p C lowast) oplusH p +q (F p C lowast)
oplusH p +q (F p C lowastF p minus1C lowast)
i
j
k
which gives us a spectral sequence E r pq
From now on we concentrate on spectral sequences arising from an exact couple which
complies the following two conditions we require D pq = 0 if p lt 0 and for q lt 0 we want
i D p minus1q +1 D pq to be an isomorphism
Remark ose two facts guarantee that the spectral sequence lives in the first quadrant
Proof First let p lt 0 Since k E pq D p minus1q = 0 has im(k ) = 0 we get d = jk = 0
Furthermore 0 = D pq j
E pq says that
E pq = ker(d )983087
im(d ) = ker(d ) = im( j ) = 0
Now we look at the case q lt 0 We have d E pq D p minus1q
j
D p minus1q and we know
that i D p minus1q sim D p q minus1 is an isomorphism But since im(k ) = ker(i ) = 0 this leads to
d = jk = 0 Moreover we have D p minus1q +1
i
sim D pq j
E pq and since ker( j ) = im(i ) =D p minus1q +1 we get
E pq = ker(d ) = ker(k ) = im( j ) = 0
Let Z r pq = ker(d r pq ) and B r pq = im(d r p +r q minusr +1) be the cycles and boundaries given by the
fact that the r -sheet of the spectral sequence contains chain complexes
Furthermore we set H n = limminusminusrarr
D p nminus p (the injective limit along the maps i D pq
D p +1q minus1 ie the disjoint union of the D p nminus p where two elements are said to be equal if
they are equal under compositions of i ) and F p H n = im(D pq H p +q ) by what we get
a filtration
F p minus1H n sube F p H n sube
since i (D p minus1nminus p minus1) sube D p nminus p is filtration is even canonically bounded D p nminus p = 0for p lt 0 causes F p H n = 0 If p gt n (which is equal to q = n minus p lt 0) we know that
D p n minus p D p +1nminus( p +1) and therefore H p +q = im(D p nminus p H p +q ) = F p H n
In particular the filtration satisfiescup
p F p H n = H n = limlarrminusminus
(H n F p H n) andcap
p F p H n = 0
So if a spectral sequence weakly converges to H lowast it even converges to H lowast
Proposition ere is a natural inclusion of F p H n F p minus1H n in E infin p nminus p e spectral se-
quence E r pq weakly converges to H lowast if and only if
Z infin =991785r
k minus1(i r D ) equals k minus1(0) = j (D )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 919
Proof First we define K pq to be the kernel of D pq H p +q Beeing an element of K pq is
equivalent to lying in the kernel of i r for an integer r so K pq =cup
r ker(i r ) It follows that
j (K pq ) =cup
r j (ker(i r )) =cup
r B r pq = B infin pq as one can easily see that the infinite boundaries
are the union of the B r pq (and similarly Z infin pq =cap
r Z r pq ) We look at the following diagram
0
0 K p minus1q +1 D p minusq q +1 F p minus1H n
0
0 K pq D pq F p H n 0
coker(i |K p minus1q +1)
coker(i ) F p H nF p minus1H n
i |K p minus1q +1 i
with short exact sequences as rows
Before applying the snake lemma we regard the cokernels
D pq 983087i (D p minus1q +1) = D pq
983087im(i ) = D pq
983087ker( j )
tells us that coker(i ) = j (D pq ) Furthermore we have im(i |K p minus1q +1) = ker( j |K pq ) () and
therefore coker(i |K p minus1q +1) j (K pq ) = B infin pq e snake lemma now says that
0 B infin pq j (D pq ) F p H n
983087F p minus1H n
0
is a short exact sequence
For all r isin Z we have
j (D pq ) = im( j ) = ker(k ) = k minus1(0) sube k minus1(i r D p minusr minus1q +r ) = Z r pq
what means j (D pq ) subecap
r Z r pq = Z infin pq We get a natural inclusion
F p H n983087F p minusq H n
j (D pq )1048623B infin pq sube Z infin pq
1048623B infin pq = E infin pq
and in particular we know now that the spectral sequences weakly converges to H lowast ie
F p H n F p minus1H n = E infin pq if and only if j (D ) = Z infin
Corollary (Convergence eorem) A spectral sequence with the requirements mentioned
above converges to H lowast = limminusminusrarr D E r pq rArr H p +q
If a spectral sequence arises from an exact couple of a complex C lowast with canonically bounded
filtration it converges to limminusminusrarr
D = H lowast which is the homology of the filtered complex C lowast
Proof A remark above tells us that the spectral sequence converges to H lowast if it weakly con-
verges to H lowast All we have to show is that Z infin = k minus1(0) if the spectral sequence complies
the two facts that D pq = 0 for p lt 0 and i D p minus1q +1
sim D pq for q lt 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1019
Let p and q be arbitrary integers ere exists an integer r such that p minus r minus 1 lt 0 which
implies i r (D p minusr minus1q +r ) = i r (0) = 0 We now see that Z r pq = k minus1(i r (D p minusr minus1q +r ) =
k minus1(0) en Z infin pq =cap
r prime Z r prime
pq = k minus1(0) and therefore Z infin = k minus1(0)
at means E r pq rArr H p +q
Let F p minus1C lowast sube F p C lowast sube sube C lowast be a canonically bounded filtration of a complex C lowastSince F p C n = 0 for all p lt 0 we get D pq = H p +q (F p C lowast) = 0 for those p For q lt 0 we have
p = n minusq gt n and F p C n = C n is leads to D pq = H n(F p C lowast) = H n(C lowast) = H n(F p +1C lowast) =D p +1q minus1 = D p +1q minus2 Alltogether it follows that a spectral sequence arising from a
canonically bounded complex converges to H lowast
It remains to show hat H lowast is the homology of C lowast
Let p lt n we then know that i is an isomorphism and therefore we get
H p +q im(D p nminus p H p +q ) D p nminus p = H p +q (F p C lowast) = H p +q (C lowast)
3 Double ComplexesIn Example we explained that a filtered complex yields an exact couple which in turn
yields a spectral sequence We will now go one step further and introduce double complexes
which yield filtered complexes and finally spectral sequences
Definition (Double complex) A double complex C lowastlowast is a family (C pq ) p q isinZ of objects
of an abelian category together with maps
d h pq C pq C p minus1q d v pq C pq C p q minus1
such that for each q resp p the families (C lowastq d hlowastq ) resp (C p lowast d v p lowast) are chain complexes
and additionally in every square the relation
d v p minus1q d h pq = plusmn d h p q minus1 d v pq
is satisfied In case of a ldquo+rdquo in this relation we call the double complex commutative and
in case of a ldquominusrdquo we call it anticommutative We say that a double complex lives in the first
quadrant if C pq = 0 whenever p lt 0 or q lt 0
We can picture a double complex like this (here we have a first quadrant double complex)
C 02
C 12
C 22
C 01 C 11 C 21
C 00 C 10 C 20
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1119
Definition Let C lowastlowast be an anticommutative double complex en we define the total
complex aached to C lowastlowast as the chain complex Totlowast(C lowastlowast) with entries⁴
Totn(C lowastlowast) =991893
p +q =n
C pq
and differentials d = d v + d h
A straightforward calculation using anticommutativity shows d 2 = 0 so that the total
complex is indeed a chain complex
We can turn the total complex of a given double complex into a filtered chain complex by
considering the following two obvious filtrations
ldquoFiltration by rowsrdquo
0 0 0
lowast lowast lowast
lowast lowast lowast
ldquoFiltration by columnsrdquo
lowast lowast 0lowast lowast 0lowast lowast 0
More precisely we set
F (R) p Totn(C lowastlowast) =
991893i le p
C i nminusi F (C)q Totn(C lowastlowast) =
991893 j leq
C nminus j j
In both cases we get
983080F p Totlowast(C lowastlowast)
983087F p minus1 Totlowast(C lowastlowast)
983081n
= C pq
for the n-th entry of the filtration quotients (where p + q = n) Note that when the double
complex lives in the first quadrant then both filtrations are canonically bounded ereforewe will assume from now on that every double complex lives in the first quadrant
e machinery developped in the previous sections now yields two spectral sequences
which we call (R)E r lowastlowast and (C)E r lowastlowast respectively By Corollary both spectral sequences
converge to the homology of the total complex Totlowast(C lowastlowast) Oen one is not really inter-
ested in this homology but one can play these sequences off against one another to obtain
information about the entries of the spectral sequence
Can we write down more explicitly how these spectral sequences look like We have to
apply the technique of exact couples to our situation Doing so we obtain
Proposition e E 1 page of (C)E r lowastlowast is obtained from the initial double complex C lowastlowast by
taking homology in vertical direction and the differentials d 1 on that page are the maps in-
duced by the horizontal differentials d h of C lowastlowast on this homology Dually the E 1 page of (R)E r lowastlowastis obtained by ldquotransposingrdquo this procedure
⁴ It is also common to consider a slightly differently defined total complex in which the direct sum is replacedby a direct product As we will mostly consider double complexes living in the first quadrant this does notmake a difference for us
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1219
Idea of proof To prove this we would have to go into the details of how the long homology
sequence coming from the short exact sequence of chain complexes
0 F p minus1 Totlowast(C lowastlowast) F p Totlowast(C lowastlowast) F p Totlowast(C lowastlowast)1048623F p minus1 Totlowast(C lowastlowast) 0
is constructed which in the end boils down to investigating the definition of the connectinghomomorphism coming from the snake lemma As this is a rather tedious calculation we
will omit this here
is proposition suggests to view the initial double complex C lowastlowast as the E 0 page of either
spectral sequence (possibly ldquotransposedrdquo)
Before giving some examples of applications of this construction we record the following
useful result
Lemma If we have a converging spectral sequence
E r pq rArr H p +q
and the E 2 page consists of 0 everywhere except in the boom row then we have
E 2 p 0 = H p
Proof is follows directly from Remark
Finally we note that we can also apply this whole technique to a commutative double com-
plex We can turn any such double complex into an anticommutative one by inserting some
minus1rsquos for example at every second horizontal or vertical arrow in this way in every square
exactly one map is multiplied by minus1 is does not affect the remaining properties of being
a double complex and does also not change homology
4 Examples
We conclude this note with a few examples that demonstrate the power of our techniques
e first examples are very simple whereas the later ones are a bit more interesting
41 The five lemma
As a first example we want to prove the five lemma Given have a commutative diagram
A B C D E
Aprime B prime C prime D prime E prime
f
д
f prime
д prime
α β γ δ ε
with exact rows we want to prove
(a) When β δ are monomorphisms and α is an epimorphism then γ is a monomorphism
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 819
where i and j are the maps induces by i and π p above and δ is the map delivered by the
snake lemma
Now we can roll up those long exact sequences into an exact triangle
oplusH p +q (F p C lowast) oplusH p +q (F p C lowast)
oplusH p +q (F p C lowastF p minus1C lowast)
i
j
k
which gives us a spectral sequence E r pq
From now on we concentrate on spectral sequences arising from an exact couple which
complies the following two conditions we require D pq = 0 if p lt 0 and for q lt 0 we want
i D p minus1q +1 D pq to be an isomorphism
Remark ose two facts guarantee that the spectral sequence lives in the first quadrant
Proof First let p lt 0 Since k E pq D p minus1q = 0 has im(k ) = 0 we get d = jk = 0
Furthermore 0 = D pq j
E pq says that
E pq = ker(d )983087
im(d ) = ker(d ) = im( j ) = 0
Now we look at the case q lt 0 We have d E pq D p minus1q
j
D p minus1q and we know
that i D p minus1q sim D p q minus1 is an isomorphism But since im(k ) = ker(i ) = 0 this leads to
d = jk = 0 Moreover we have D p minus1q +1
i
sim D pq j
E pq and since ker( j ) = im(i ) =D p minus1q +1 we get
E pq = ker(d ) = ker(k ) = im( j ) = 0
Let Z r pq = ker(d r pq ) and B r pq = im(d r p +r q minusr +1) be the cycles and boundaries given by the
fact that the r -sheet of the spectral sequence contains chain complexes
Furthermore we set H n = limminusminusrarr
D p nminus p (the injective limit along the maps i D pq
D p +1q minus1 ie the disjoint union of the D p nminus p where two elements are said to be equal if
they are equal under compositions of i ) and F p H n = im(D pq H p +q ) by what we get
a filtration
F p minus1H n sube F p H n sube
since i (D p minus1nminus p minus1) sube D p nminus p is filtration is even canonically bounded D p nminus p = 0for p lt 0 causes F p H n = 0 If p gt n (which is equal to q = n minus p lt 0) we know that
D p n minus p D p +1nminus( p +1) and therefore H p +q = im(D p nminus p H p +q ) = F p H n
In particular the filtration satisfiescup
p F p H n = H n = limlarrminusminus
(H n F p H n) andcap
p F p H n = 0
So if a spectral sequence weakly converges to H lowast it even converges to H lowast
Proposition ere is a natural inclusion of F p H n F p minus1H n in E infin p nminus p e spectral se-
quence E r pq weakly converges to H lowast if and only if
Z infin =991785r
k minus1(i r D ) equals k minus1(0) = j (D )
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 919
Proof First we define K pq to be the kernel of D pq H p +q Beeing an element of K pq is
equivalent to lying in the kernel of i r for an integer r so K pq =cup
r ker(i r ) It follows that
j (K pq ) =cup
r j (ker(i r )) =cup
r B r pq = B infin pq as one can easily see that the infinite boundaries
are the union of the B r pq (and similarly Z infin pq =cap
r Z r pq ) We look at the following diagram
0
0 K p minus1q +1 D p minusq q +1 F p minus1H n
0
0 K pq D pq F p H n 0
coker(i |K p minus1q +1)
coker(i ) F p H nF p minus1H n
i |K p minus1q +1 i
with short exact sequences as rows
Before applying the snake lemma we regard the cokernels
D pq 983087i (D p minus1q +1) = D pq
983087im(i ) = D pq
983087ker( j )
tells us that coker(i ) = j (D pq ) Furthermore we have im(i |K p minus1q +1) = ker( j |K pq ) () and
therefore coker(i |K p minus1q +1) j (K pq ) = B infin pq e snake lemma now says that
0 B infin pq j (D pq ) F p H n
983087F p minus1H n
0
is a short exact sequence
For all r isin Z we have
j (D pq ) = im( j ) = ker(k ) = k minus1(0) sube k minus1(i r D p minusr minus1q +r ) = Z r pq
what means j (D pq ) subecap
r Z r pq = Z infin pq We get a natural inclusion
F p H n983087F p minusq H n
j (D pq )1048623B infin pq sube Z infin pq
1048623B infin pq = E infin pq
and in particular we know now that the spectral sequences weakly converges to H lowast ie
F p H n F p minus1H n = E infin pq if and only if j (D ) = Z infin
Corollary (Convergence eorem) A spectral sequence with the requirements mentioned
above converges to H lowast = limminusminusrarr D E r pq rArr H p +q
If a spectral sequence arises from an exact couple of a complex C lowast with canonically bounded
filtration it converges to limminusminusrarr
D = H lowast which is the homology of the filtered complex C lowast
Proof A remark above tells us that the spectral sequence converges to H lowast if it weakly con-
verges to H lowast All we have to show is that Z infin = k minus1(0) if the spectral sequence complies
the two facts that D pq = 0 for p lt 0 and i D p minus1q +1
sim D pq for q lt 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1019
Let p and q be arbitrary integers ere exists an integer r such that p minus r minus 1 lt 0 which
implies i r (D p minusr minus1q +r ) = i r (0) = 0 We now see that Z r pq = k minus1(i r (D p minusr minus1q +r ) =
k minus1(0) en Z infin pq =cap
r prime Z r prime
pq = k minus1(0) and therefore Z infin = k minus1(0)
at means E r pq rArr H p +q
Let F p minus1C lowast sube F p C lowast sube sube C lowast be a canonically bounded filtration of a complex C lowastSince F p C n = 0 for all p lt 0 we get D pq = H p +q (F p C lowast) = 0 for those p For q lt 0 we have
p = n minusq gt n and F p C n = C n is leads to D pq = H n(F p C lowast) = H n(C lowast) = H n(F p +1C lowast) =D p +1q minus1 = D p +1q minus2 Alltogether it follows that a spectral sequence arising from a
canonically bounded complex converges to H lowast
It remains to show hat H lowast is the homology of C lowast
Let p lt n we then know that i is an isomorphism and therefore we get
H p +q im(D p nminus p H p +q ) D p nminus p = H p +q (F p C lowast) = H p +q (C lowast)
3 Double ComplexesIn Example we explained that a filtered complex yields an exact couple which in turn
yields a spectral sequence We will now go one step further and introduce double complexes
which yield filtered complexes and finally spectral sequences
Definition (Double complex) A double complex C lowastlowast is a family (C pq ) p q isinZ of objects
of an abelian category together with maps
d h pq C pq C p minus1q d v pq C pq C p q minus1
such that for each q resp p the families (C lowastq d hlowastq ) resp (C p lowast d v p lowast) are chain complexes
and additionally in every square the relation
d v p minus1q d h pq = plusmn d h p q minus1 d v pq
is satisfied In case of a ldquo+rdquo in this relation we call the double complex commutative and
in case of a ldquominusrdquo we call it anticommutative We say that a double complex lives in the first
quadrant if C pq = 0 whenever p lt 0 or q lt 0
We can picture a double complex like this (here we have a first quadrant double complex)
C 02
C 12
C 22
C 01 C 11 C 21
C 00 C 10 C 20
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1119
Definition Let C lowastlowast be an anticommutative double complex en we define the total
complex aached to C lowastlowast as the chain complex Totlowast(C lowastlowast) with entries⁴
Totn(C lowastlowast) =991893
p +q =n
C pq
and differentials d = d v + d h
A straightforward calculation using anticommutativity shows d 2 = 0 so that the total
complex is indeed a chain complex
We can turn the total complex of a given double complex into a filtered chain complex by
considering the following two obvious filtrations
ldquoFiltration by rowsrdquo
0 0 0
lowast lowast lowast
lowast lowast lowast
ldquoFiltration by columnsrdquo
lowast lowast 0lowast lowast 0lowast lowast 0
More precisely we set
F (R) p Totn(C lowastlowast) =
991893i le p
C i nminusi F (C)q Totn(C lowastlowast) =
991893 j leq
C nminus j j
In both cases we get
983080F p Totlowast(C lowastlowast)
983087F p minus1 Totlowast(C lowastlowast)
983081n
= C pq
for the n-th entry of the filtration quotients (where p + q = n) Note that when the double
complex lives in the first quadrant then both filtrations are canonically bounded ereforewe will assume from now on that every double complex lives in the first quadrant
e machinery developped in the previous sections now yields two spectral sequences
which we call (R)E r lowastlowast and (C)E r lowastlowast respectively By Corollary both spectral sequences
converge to the homology of the total complex Totlowast(C lowastlowast) Oen one is not really inter-
ested in this homology but one can play these sequences off against one another to obtain
information about the entries of the spectral sequence
Can we write down more explicitly how these spectral sequences look like We have to
apply the technique of exact couples to our situation Doing so we obtain
Proposition e E 1 page of (C)E r lowastlowast is obtained from the initial double complex C lowastlowast by
taking homology in vertical direction and the differentials d 1 on that page are the maps in-
duced by the horizontal differentials d h of C lowastlowast on this homology Dually the E 1 page of (R)E r lowastlowastis obtained by ldquotransposingrdquo this procedure
⁴ It is also common to consider a slightly differently defined total complex in which the direct sum is replacedby a direct product As we will mostly consider double complexes living in the first quadrant this does notmake a difference for us
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1219
Idea of proof To prove this we would have to go into the details of how the long homology
sequence coming from the short exact sequence of chain complexes
0 F p minus1 Totlowast(C lowastlowast) F p Totlowast(C lowastlowast) F p Totlowast(C lowastlowast)1048623F p minus1 Totlowast(C lowastlowast) 0
is constructed which in the end boils down to investigating the definition of the connectinghomomorphism coming from the snake lemma As this is a rather tedious calculation we
will omit this here
is proposition suggests to view the initial double complex C lowastlowast as the E 0 page of either
spectral sequence (possibly ldquotransposedrdquo)
Before giving some examples of applications of this construction we record the following
useful result
Lemma If we have a converging spectral sequence
E r pq rArr H p +q
and the E 2 page consists of 0 everywhere except in the boom row then we have
E 2 p 0 = H p
Proof is follows directly from Remark
Finally we note that we can also apply this whole technique to a commutative double com-
plex We can turn any such double complex into an anticommutative one by inserting some
minus1rsquos for example at every second horizontal or vertical arrow in this way in every square
exactly one map is multiplied by minus1 is does not affect the remaining properties of being
a double complex and does also not change homology
4 Examples
We conclude this note with a few examples that demonstrate the power of our techniques
e first examples are very simple whereas the later ones are a bit more interesting
41 The five lemma
As a first example we want to prove the five lemma Given have a commutative diagram
A B C D E
Aprime B prime C prime D prime E prime
f
д
f prime
д prime
α β γ δ ε
with exact rows we want to prove
(a) When β δ are monomorphisms and α is an epimorphism then γ is a monomorphism
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 919
Proof First we define K pq to be the kernel of D pq H p +q Beeing an element of K pq is
equivalent to lying in the kernel of i r for an integer r so K pq =cup
r ker(i r ) It follows that
j (K pq ) =cup
r j (ker(i r )) =cup
r B r pq = B infin pq as one can easily see that the infinite boundaries
are the union of the B r pq (and similarly Z infin pq =cap
r Z r pq ) We look at the following diagram
0
0 K p minus1q +1 D p minusq q +1 F p minus1H n
0
0 K pq D pq F p H n 0
coker(i |K p minus1q +1)
coker(i ) F p H nF p minus1H n
i |K p minus1q +1 i
with short exact sequences as rows
Before applying the snake lemma we regard the cokernels
D pq 983087i (D p minus1q +1) = D pq
983087im(i ) = D pq
983087ker( j )
tells us that coker(i ) = j (D pq ) Furthermore we have im(i |K p minus1q +1) = ker( j |K pq ) () and
therefore coker(i |K p minus1q +1) j (K pq ) = B infin pq e snake lemma now says that
0 B infin pq j (D pq ) F p H n
983087F p minus1H n
0
is a short exact sequence
For all r isin Z we have
j (D pq ) = im( j ) = ker(k ) = k minus1(0) sube k minus1(i r D p minusr minus1q +r ) = Z r pq
what means j (D pq ) subecap
r Z r pq = Z infin pq We get a natural inclusion
F p H n983087F p minusq H n
j (D pq )1048623B infin pq sube Z infin pq
1048623B infin pq = E infin pq
and in particular we know now that the spectral sequences weakly converges to H lowast ie
F p H n F p minus1H n = E infin pq if and only if j (D ) = Z infin
Corollary (Convergence eorem) A spectral sequence with the requirements mentioned
above converges to H lowast = limminusminusrarr D E r pq rArr H p +q
If a spectral sequence arises from an exact couple of a complex C lowast with canonically bounded
filtration it converges to limminusminusrarr
D = H lowast which is the homology of the filtered complex C lowast
Proof A remark above tells us that the spectral sequence converges to H lowast if it weakly con-
verges to H lowast All we have to show is that Z infin = k minus1(0) if the spectral sequence complies
the two facts that D pq = 0 for p lt 0 and i D p minus1q +1
sim D pq for q lt 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1019
Let p and q be arbitrary integers ere exists an integer r such that p minus r minus 1 lt 0 which
implies i r (D p minusr minus1q +r ) = i r (0) = 0 We now see that Z r pq = k minus1(i r (D p minusr minus1q +r ) =
k minus1(0) en Z infin pq =cap
r prime Z r prime
pq = k minus1(0) and therefore Z infin = k minus1(0)
at means E r pq rArr H p +q
Let F p minus1C lowast sube F p C lowast sube sube C lowast be a canonically bounded filtration of a complex C lowastSince F p C n = 0 for all p lt 0 we get D pq = H p +q (F p C lowast) = 0 for those p For q lt 0 we have
p = n minusq gt n and F p C n = C n is leads to D pq = H n(F p C lowast) = H n(C lowast) = H n(F p +1C lowast) =D p +1q minus1 = D p +1q minus2 Alltogether it follows that a spectral sequence arising from a
canonically bounded complex converges to H lowast
It remains to show hat H lowast is the homology of C lowast
Let p lt n we then know that i is an isomorphism and therefore we get
H p +q im(D p nminus p H p +q ) D p nminus p = H p +q (F p C lowast) = H p +q (C lowast)
3 Double ComplexesIn Example we explained that a filtered complex yields an exact couple which in turn
yields a spectral sequence We will now go one step further and introduce double complexes
which yield filtered complexes and finally spectral sequences
Definition (Double complex) A double complex C lowastlowast is a family (C pq ) p q isinZ of objects
of an abelian category together with maps
d h pq C pq C p minus1q d v pq C pq C p q minus1
such that for each q resp p the families (C lowastq d hlowastq ) resp (C p lowast d v p lowast) are chain complexes
and additionally in every square the relation
d v p minus1q d h pq = plusmn d h p q minus1 d v pq
is satisfied In case of a ldquo+rdquo in this relation we call the double complex commutative and
in case of a ldquominusrdquo we call it anticommutative We say that a double complex lives in the first
quadrant if C pq = 0 whenever p lt 0 or q lt 0
We can picture a double complex like this (here we have a first quadrant double complex)
C 02
C 12
C 22
C 01 C 11 C 21
C 00 C 10 C 20
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1119
Definition Let C lowastlowast be an anticommutative double complex en we define the total
complex aached to C lowastlowast as the chain complex Totlowast(C lowastlowast) with entries⁴
Totn(C lowastlowast) =991893
p +q =n
C pq
and differentials d = d v + d h
A straightforward calculation using anticommutativity shows d 2 = 0 so that the total
complex is indeed a chain complex
We can turn the total complex of a given double complex into a filtered chain complex by
considering the following two obvious filtrations
ldquoFiltration by rowsrdquo
0 0 0
lowast lowast lowast
lowast lowast lowast
ldquoFiltration by columnsrdquo
lowast lowast 0lowast lowast 0lowast lowast 0
More precisely we set
F (R) p Totn(C lowastlowast) =
991893i le p
C i nminusi F (C)q Totn(C lowastlowast) =
991893 j leq
C nminus j j
In both cases we get
983080F p Totlowast(C lowastlowast)
983087F p minus1 Totlowast(C lowastlowast)
983081n
= C pq
for the n-th entry of the filtration quotients (where p + q = n) Note that when the double
complex lives in the first quadrant then both filtrations are canonically bounded ereforewe will assume from now on that every double complex lives in the first quadrant
e machinery developped in the previous sections now yields two spectral sequences
which we call (R)E r lowastlowast and (C)E r lowastlowast respectively By Corollary both spectral sequences
converge to the homology of the total complex Totlowast(C lowastlowast) Oen one is not really inter-
ested in this homology but one can play these sequences off against one another to obtain
information about the entries of the spectral sequence
Can we write down more explicitly how these spectral sequences look like We have to
apply the technique of exact couples to our situation Doing so we obtain
Proposition e E 1 page of (C)E r lowastlowast is obtained from the initial double complex C lowastlowast by
taking homology in vertical direction and the differentials d 1 on that page are the maps in-
duced by the horizontal differentials d h of C lowastlowast on this homology Dually the E 1 page of (R)E r lowastlowastis obtained by ldquotransposingrdquo this procedure
⁴ It is also common to consider a slightly differently defined total complex in which the direct sum is replacedby a direct product As we will mostly consider double complexes living in the first quadrant this does notmake a difference for us
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1219
Idea of proof To prove this we would have to go into the details of how the long homology
sequence coming from the short exact sequence of chain complexes
0 F p minus1 Totlowast(C lowastlowast) F p Totlowast(C lowastlowast) F p Totlowast(C lowastlowast)1048623F p minus1 Totlowast(C lowastlowast) 0
is constructed which in the end boils down to investigating the definition of the connectinghomomorphism coming from the snake lemma As this is a rather tedious calculation we
will omit this here
is proposition suggests to view the initial double complex C lowastlowast as the E 0 page of either
spectral sequence (possibly ldquotransposedrdquo)
Before giving some examples of applications of this construction we record the following
useful result
Lemma If we have a converging spectral sequence
E r pq rArr H p +q
and the E 2 page consists of 0 everywhere except in the boom row then we have
E 2 p 0 = H p
Proof is follows directly from Remark
Finally we note that we can also apply this whole technique to a commutative double com-
plex We can turn any such double complex into an anticommutative one by inserting some
minus1rsquos for example at every second horizontal or vertical arrow in this way in every square
exactly one map is multiplied by minus1 is does not affect the remaining properties of being
a double complex and does also not change homology
4 Examples
We conclude this note with a few examples that demonstrate the power of our techniques
e first examples are very simple whereas the later ones are a bit more interesting
41 The five lemma
As a first example we want to prove the five lemma Given have a commutative diagram
A B C D E
Aprime B prime C prime D prime E prime
f
д
f prime
д prime
α β γ δ ε
with exact rows we want to prove
(a) When β δ are monomorphisms and α is an epimorphism then γ is a monomorphism
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1019
Let p and q be arbitrary integers ere exists an integer r such that p minus r minus 1 lt 0 which
implies i r (D p minusr minus1q +r ) = i r (0) = 0 We now see that Z r pq = k minus1(i r (D p minusr minus1q +r ) =
k minus1(0) en Z infin pq =cap
r prime Z r prime
pq = k minus1(0) and therefore Z infin = k minus1(0)
at means E r pq rArr H p +q
Let F p minus1C lowast sube F p C lowast sube sube C lowast be a canonically bounded filtration of a complex C lowastSince F p C n = 0 for all p lt 0 we get D pq = H p +q (F p C lowast) = 0 for those p For q lt 0 we have
p = n minusq gt n and F p C n = C n is leads to D pq = H n(F p C lowast) = H n(C lowast) = H n(F p +1C lowast) =D p +1q minus1 = D p +1q minus2 Alltogether it follows that a spectral sequence arising from a
canonically bounded complex converges to H lowast
It remains to show hat H lowast is the homology of C lowast
Let p lt n we then know that i is an isomorphism and therefore we get
H p +q im(D p nminus p H p +q ) D p nminus p = H p +q (F p C lowast) = H p +q (C lowast)
3 Double ComplexesIn Example we explained that a filtered complex yields an exact couple which in turn
yields a spectral sequence We will now go one step further and introduce double complexes
which yield filtered complexes and finally spectral sequences
Definition (Double complex) A double complex C lowastlowast is a family (C pq ) p q isinZ of objects
of an abelian category together with maps
d h pq C pq C p minus1q d v pq C pq C p q minus1
such that for each q resp p the families (C lowastq d hlowastq ) resp (C p lowast d v p lowast) are chain complexes
and additionally in every square the relation
d v p minus1q d h pq = plusmn d h p q minus1 d v pq
is satisfied In case of a ldquo+rdquo in this relation we call the double complex commutative and
in case of a ldquominusrdquo we call it anticommutative We say that a double complex lives in the first
quadrant if C pq = 0 whenever p lt 0 or q lt 0
We can picture a double complex like this (here we have a first quadrant double complex)
C 02
C 12
C 22
C 01 C 11 C 21
C 00 C 10 C 20
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1119
Definition Let C lowastlowast be an anticommutative double complex en we define the total
complex aached to C lowastlowast as the chain complex Totlowast(C lowastlowast) with entries⁴
Totn(C lowastlowast) =991893
p +q =n
C pq
and differentials d = d v + d h
A straightforward calculation using anticommutativity shows d 2 = 0 so that the total
complex is indeed a chain complex
We can turn the total complex of a given double complex into a filtered chain complex by
considering the following two obvious filtrations
ldquoFiltration by rowsrdquo
0 0 0
lowast lowast lowast
lowast lowast lowast
ldquoFiltration by columnsrdquo
lowast lowast 0lowast lowast 0lowast lowast 0
More precisely we set
F (R) p Totn(C lowastlowast) =
991893i le p
C i nminusi F (C)q Totn(C lowastlowast) =
991893 j leq
C nminus j j
In both cases we get
983080F p Totlowast(C lowastlowast)
983087F p minus1 Totlowast(C lowastlowast)
983081n
= C pq
for the n-th entry of the filtration quotients (where p + q = n) Note that when the double
complex lives in the first quadrant then both filtrations are canonically bounded ereforewe will assume from now on that every double complex lives in the first quadrant
e machinery developped in the previous sections now yields two spectral sequences
which we call (R)E r lowastlowast and (C)E r lowastlowast respectively By Corollary both spectral sequences
converge to the homology of the total complex Totlowast(C lowastlowast) Oen one is not really inter-
ested in this homology but one can play these sequences off against one another to obtain
information about the entries of the spectral sequence
Can we write down more explicitly how these spectral sequences look like We have to
apply the technique of exact couples to our situation Doing so we obtain
Proposition e E 1 page of (C)E r lowastlowast is obtained from the initial double complex C lowastlowast by
taking homology in vertical direction and the differentials d 1 on that page are the maps in-
duced by the horizontal differentials d h of C lowastlowast on this homology Dually the E 1 page of (R)E r lowastlowastis obtained by ldquotransposingrdquo this procedure
⁴ It is also common to consider a slightly differently defined total complex in which the direct sum is replacedby a direct product As we will mostly consider double complexes living in the first quadrant this does notmake a difference for us
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1219
Idea of proof To prove this we would have to go into the details of how the long homology
sequence coming from the short exact sequence of chain complexes
0 F p minus1 Totlowast(C lowastlowast) F p Totlowast(C lowastlowast) F p Totlowast(C lowastlowast)1048623F p minus1 Totlowast(C lowastlowast) 0
is constructed which in the end boils down to investigating the definition of the connectinghomomorphism coming from the snake lemma As this is a rather tedious calculation we
will omit this here
is proposition suggests to view the initial double complex C lowastlowast as the E 0 page of either
spectral sequence (possibly ldquotransposedrdquo)
Before giving some examples of applications of this construction we record the following
useful result
Lemma If we have a converging spectral sequence
E r pq rArr H p +q
and the E 2 page consists of 0 everywhere except in the boom row then we have
E 2 p 0 = H p
Proof is follows directly from Remark
Finally we note that we can also apply this whole technique to a commutative double com-
plex We can turn any such double complex into an anticommutative one by inserting some
minus1rsquos for example at every second horizontal or vertical arrow in this way in every square
exactly one map is multiplied by minus1 is does not affect the remaining properties of being
a double complex and does also not change homology
4 Examples
We conclude this note with a few examples that demonstrate the power of our techniques
e first examples are very simple whereas the later ones are a bit more interesting
41 The five lemma
As a first example we want to prove the five lemma Given have a commutative diagram
A B C D E
Aprime B prime C prime D prime E prime
f
д
f prime
д prime
α β γ δ ε
with exact rows we want to prove
(a) When β δ are monomorphisms and α is an epimorphism then γ is a monomorphism
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1119
Definition Let C lowastlowast be an anticommutative double complex en we define the total
complex aached to C lowastlowast as the chain complex Totlowast(C lowastlowast) with entries⁴
Totn(C lowastlowast) =991893
p +q =n
C pq
and differentials d = d v + d h
A straightforward calculation using anticommutativity shows d 2 = 0 so that the total
complex is indeed a chain complex
We can turn the total complex of a given double complex into a filtered chain complex by
considering the following two obvious filtrations
ldquoFiltration by rowsrdquo
0 0 0
lowast lowast lowast
lowast lowast lowast
ldquoFiltration by columnsrdquo
lowast lowast 0lowast lowast 0lowast lowast 0
More precisely we set
F (R) p Totn(C lowastlowast) =
991893i le p
C i nminusi F (C)q Totn(C lowastlowast) =
991893 j leq
C nminus j j
In both cases we get
983080F p Totlowast(C lowastlowast)
983087F p minus1 Totlowast(C lowastlowast)
983081n
= C pq
for the n-th entry of the filtration quotients (where p + q = n) Note that when the double
complex lives in the first quadrant then both filtrations are canonically bounded ereforewe will assume from now on that every double complex lives in the first quadrant
e machinery developped in the previous sections now yields two spectral sequences
which we call (R)E r lowastlowast and (C)E r lowastlowast respectively By Corollary both spectral sequences
converge to the homology of the total complex Totlowast(C lowastlowast) Oen one is not really inter-
ested in this homology but one can play these sequences off against one another to obtain
information about the entries of the spectral sequence
Can we write down more explicitly how these spectral sequences look like We have to
apply the technique of exact couples to our situation Doing so we obtain
Proposition e E 1 page of (C)E r lowastlowast is obtained from the initial double complex C lowastlowast by
taking homology in vertical direction and the differentials d 1 on that page are the maps in-
duced by the horizontal differentials d h of C lowastlowast on this homology Dually the E 1 page of (R)E r lowastlowastis obtained by ldquotransposingrdquo this procedure
⁴ It is also common to consider a slightly differently defined total complex in which the direct sum is replacedby a direct product As we will mostly consider double complexes living in the first quadrant this does notmake a difference for us
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1219
Idea of proof To prove this we would have to go into the details of how the long homology
sequence coming from the short exact sequence of chain complexes
0 F p minus1 Totlowast(C lowastlowast) F p Totlowast(C lowastlowast) F p Totlowast(C lowastlowast)1048623F p minus1 Totlowast(C lowastlowast) 0
is constructed which in the end boils down to investigating the definition of the connectinghomomorphism coming from the snake lemma As this is a rather tedious calculation we
will omit this here
is proposition suggests to view the initial double complex C lowastlowast as the E 0 page of either
spectral sequence (possibly ldquotransposedrdquo)
Before giving some examples of applications of this construction we record the following
useful result
Lemma If we have a converging spectral sequence
E r pq rArr H p +q
and the E 2 page consists of 0 everywhere except in the boom row then we have
E 2 p 0 = H p
Proof is follows directly from Remark
Finally we note that we can also apply this whole technique to a commutative double com-
plex We can turn any such double complex into an anticommutative one by inserting some
minus1rsquos for example at every second horizontal or vertical arrow in this way in every square
exactly one map is multiplied by minus1 is does not affect the remaining properties of being
a double complex and does also not change homology
4 Examples
We conclude this note with a few examples that demonstrate the power of our techniques
e first examples are very simple whereas the later ones are a bit more interesting
41 The five lemma
As a first example we want to prove the five lemma Given have a commutative diagram
A B C D E
Aprime B prime C prime D prime E prime
f
д
f prime
д prime
α β γ δ ε
with exact rows we want to prove
(a) When β δ are monomorphisms and α is an epimorphism then γ is a monomorphism
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1219
Idea of proof To prove this we would have to go into the details of how the long homology
sequence coming from the short exact sequence of chain complexes
0 F p minus1 Totlowast(C lowastlowast) F p Totlowast(C lowastlowast) F p Totlowast(C lowastlowast)1048623F p minus1 Totlowast(C lowastlowast) 0
is constructed which in the end boils down to investigating the definition of the connectinghomomorphism coming from the snake lemma As this is a rather tedious calculation we
will omit this here
is proposition suggests to view the initial double complex C lowastlowast as the E 0 page of either
spectral sequence (possibly ldquotransposedrdquo)
Before giving some examples of applications of this construction we record the following
useful result
Lemma If we have a converging spectral sequence
E r pq rArr H p +q
and the E 2 page consists of 0 everywhere except in the boom row then we have
E 2 p 0 = H p
Proof is follows directly from Remark
Finally we note that we can also apply this whole technique to a commutative double com-
plex We can turn any such double complex into an anticommutative one by inserting some
minus1rsquos for example at every second horizontal or vertical arrow in this way in every square
exactly one map is multiplied by minus1 is does not affect the remaining properties of being
a double complex and does also not change homology
4 Examples
We conclude this note with a few examples that demonstrate the power of our techniques
e first examples are very simple whereas the later ones are a bit more interesting
41 The five lemma
As a first example we want to prove the five lemma Given have a commutative diagram
A B C D E
Aprime B prime C prime D prime E prime
f
д
f prime
д prime
α β γ δ ε
with exact rows we want to prove
(a) When β δ are monomorphisms and α is an epimorphism then γ is a monomorphism
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1319
(b) When β δ are epimorphisms and ε is a monomorphism then γ is an epimorphism
To get a double complex we flip this diagram insert kernels and cokernels on the le and
right and fill everything else with 0 to obtain
0 0 0 0 0 0 0 0
coker д E D C B A
ker f
0
coker дprime E prime D prime C prime B prime Aprime ker f prime 0
In the following we will omit the 0rsquos again and draw only the non-zero part
From this we see immediately that the E 1 page of (R)E r lowastlowast consists only of 0 because to get
this E 1 page we have to take homology in horizontal direction but as the rows are exactthis vanishes So we conclude that (R)E r lowastlowast converges to 0 and our theory then tells us that
also (C)E r lowastlowast must converge to 0 What does this imply
Taking homology in vertical direction we obtain the E 1 page of (C)E r lowastlowast
lowast ker ε ker δ kerγ ker β ker α lowast
lowast
coker ε
coker δ
cokerγ
coker β
coker α
lowast
Here we have wrien ldquolowastrdquo to indicate any entry we are not interested in
We now use the assumption from (a) to get ker δ = ker β = cokerα = 0 By taking
homology again we finally look at the E 2 page
lowast
lowast
lowast
kerγ
lowast
lowast
lowast
lowast lowast lowast lowast lowast 0 lowast
As the spectral sequence converges to 0 we know that on the E infin page no entry can be
le But this means that kerγ has to vanish since otherwise it could never disappear on
the following pages is is just what we wanted to show in (a) e claim in (b) is shown
similarly
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1419
42 The snake lemma
e next application of our theory will be a proof of the snake lemma which proceeds quite
similar to the one of the five lemma We start again with a commutative diagram
A
B
C
0
0 Aprime B prime C prime
f
д
f prime
д prime
α β γ
with exact rows and as before we insert kernels cokernels and zeros to obtain the E 0 page
of our two spectral sequences
0 0 0 0 0 0 0
0 0 C B A ker f 0
0
coker дprime C prime B prime Aprime
0
0
As the rows are exact horizontal homology vanishes and again both spectral sequences
converge to 0 Now taking homology in vertical direction yields the E 1 page of (C)E r lowastlowast
0
kerγ
ker β
ker α
ker f
0
0
cokerдprime
cokerγ
coker β
cokerα
0
φ
ψ
Here we see part of the snake lemma sequence in the two rows and standard arguments
show that these are exact Finally we take homology again to look at the E 2 page
0 cokerφ 0 0 0
0
0
0
kerψ
0
As on the E infin page everything must have vanished this one remaining map must be an
isomorphism By inverting this isomorphism we can get a connecting homomorphism andobtain the snake lemma sequence
0
ker f
kerα
ker β
kerγ
cokerα
coker β
cokerγ
coker дprime 0
cokerφ kerψ
φ
ψ
sim
By simple arguments one can prove that this is indeed exact
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1519
43 Balancing Tor
e next application we want to consider is balancing Tor So let R be a ring A a right
R -module and B a le R -module Denoting by Torlowast(A minus) the derived functors of A otimesR minus
and by Torlowast(minus B ) the derived functors of minus otimesR B we want to show
Torlowast(A minus)(B ) = Torlowast(minus B )(A)
using spectral sequences
To get our machinery started we need a double complex We choose projective resolutions
P 2 P 1 P 0 A 0
Q 2 Q 1 Q 0 B 0
of A and B respectively and define the double complex C lowastlowast by
C pq = P p otimes Q q
where the maps are the induced ones coming from the maps in the projective resolutions
Since projective modules are flat the rows and columns of this double complex are indeed
complexes and the squares in the double complex are commutative
Writing down the E 0 page of the column-filtered spectral sequence we have
P 0 otimes Q 2 P 1 otimes Q 2 P 2 otimes Q 2
P 0 otimes Q 1
P 1 otimes Q 1
P 2 otimes Q 1
P 0 otimes Q 0 P 1 otimes Q 0 P 2 otimes Q 0
Now since every P p is projective (hence flat) the sequence
P p otimes Q 1 P p otimes Q 0 P p otimes B 0
is exact so we have
coker(P p otimes Q 1 P p otimes Q 0) = P p otimes B
erefore the E 1 page looks like this
0
0
0
P 0 otimes B P 1 otimes B P 2 otimes B
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1619
What we have here in the boom row is exactly the complex used to calculate the derived
functors of minus otimes B So taking homology of this yields as E 2 page of (C)E r lowastlowast
0 0 0 0 0
Tor0(minus B )(A)
Tor1(minus B )(A)
Tor2(minus B )(A)
Tor3(minus B )(A)
Tor4(minus B )(A)
In a totally analogous way one can see that the E 2 page of (R)E r lowastlowast is
0 0 0 0 0
Tor0(A minus)(B )
Tor1(A minus)(B )
Tor2(A minus)(B )
Tor3(A minus)(B )
Tor4(A minus)(B )
Now in both spectral sequences we have an E 2 page whose only nonzero entries are in the
boom row In this situation we can apply Lemma which tells us that
(C)E r pq rArr Tor p +q (minus B )(A) and (R)E r pq rArr Tor p +q (A minus)(B )
But as both spectral sequences converge to the same thing we are done now
44 Base change for Tor
We now consider the following situation let R S be rings f R S a ring homomor-
phism A an R -module and B an S -module By f we can also consider B as an R -module
As in the previous example we consider again projective resolutions
P 2 P 1 P 0 A
0
Q 2
Q 1
Q 0
B 0
of A as an R -module and B as an S -module respectively and form the double complex C lowastlowastgiven by
C pq = P p otimesR Q q
Here Q q might not be projective as an R -module But P p is so P p otimesR minus is exact and when
we consider the spectral sequence (C)E r lowastlowast filtered by columns we see by the same argument
as an the previous example that it converges to TorR p +q (A B ) So we know that also (C)E r lowastlowastconverges to the same thing but how does this spectral sequence look like explicitly
e E 0 page looks like this
P 2 otimesR Q 0 P 2 otimesR Q 1 P 2 otimesR Q 2
P 1 otimesR Q 0 P 1 otimesR Q 1 P 1 otimesR Q 2
P 0 otimesR Q 0 P 0 otimesR Q 1 P 0 otimesR Q 2
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1719
We now use
P p otimesR Q q = (P p otimesR S ) otimesS Q q
As (P p otimesR S ) p is just the complex used to calculate the derived functors of minus otimesR S and every
Q q is projective hence minus otimesS Q q is exact we obtain the following E 1 page
TorR 2 (A S ) otimesS Q 0
TorR 2 (A S ) otimesS Q 1
TorR 2 (A S ) otimesS Q 2
TorR 1 (A S ) otimesS Q 0 TorR 1 (A S ) otimesS Q 1 TorR 1 (A S ) otimesS Q 2
TorR 0 (A S ) otimesS Q 0
TorR 0 (A S ) otimesS Q 1
TorR 0 (A S ) otimesS Q 2
Now the complexes in these rows are the ones used to calculate the derived functors of
minus otimesS B and this means that the pq -th entry on the E 2 page is TorS p (TorR q (A S ) B ) So we
have proved
Proposition ere exists a spectral sequence
E 2 pq = TorS p (TorR q (A S ) B ) rArr TorR p +q (A B )
Why is this useful Let us consider the special case that S is flat as an R -module en
TorR i (A S ) vanishes fori gt 0 hence on the E 2 page of the spectral sequence we constructed
only the boom row contains nonzero entries erefore we can again use Lemma
which gives us
TorS i (A otimesR S B ) = TorR i (A B )
e requirement that S is a flat R -module is specifically fulfilled when G is a group H is a
subgroup and we take R = Z[H ] and S = Z[G ] In this case S is even R -free one can use
any system of coset representatives as a basis In terms of group homology (seing B = Z)
the statement then reads
H i (G indG H A) = H i (H A)
is is just Shapirorsquos lemma
45 The universal coefficients theorem
As a last application we want to prove a universal coefficients theorem Let (C lowast d lowast) =(C q d q )q be a chain complex consisting of free abelian groups and A be any abelian group
In this situation what can we say about the relation between H lowast(C lowast) and H lowast(C lowast otimes A)
We choose a projective resolution
P 2 P 1 P 0 A 0
of A and consider the double complex (P p otimes C q ) pq
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1819
We first look at the spectral sequence coming from the filtration by columns Its E 0 page is
P 0 otimes C 2 P 1 otimes C 2 P 2 otimes C 2
P 0 otimes C 1 P 1 otimes C 1 P 2 otimes C 1
P 0 otimes C 0 P 1 otimes C 0 P 2 otimes C 0
Since P p otimes minus is exact we get as E 1 page
P 0 otimes H 2(C ) P 1 otimes H 2(C ) P 2 otimes H 2(C )
P 0 otimes H 1(C ) P 1 otimes H 1(C ) P 2 otimes H 1(C )
P 0 otimes H 0(C ) P 1 otimes H 0(C ) P 2 otimes H 0(C )
Now the rows here are the complexes used to calculate the derived functors of A otimes minus so
the pq -th entry on the E 2 page is TorZq (A H p (C )) We want to examine this more closely
Applying A otimes minus to the short exact sequence of chain complexes
0 imd lowast kerd lowast H lowast(C lowast) 0
and deriving yields a long exact sequence Tor2(A imd nminus1) Tor2(A kerd n) Tor2(A H n(C ))
part Tor1(A imd nminus1) Tor1(A kerd n) Tor1(A H n(C ))
part Tor0(A imd nminus1)
for every n isin N0 But as im d nminus1 and kerd n are subgroups of the free abelian group C n
they are both free themselves and therefore the higher Tor groups vanish By the long
exact sequence this means that also Tori (A H n(C )) vanishes for i ge 2 Hence our E 2 page
looks like this
Tor0(A H 2(C )) Tor1(A H 2(C )) 0 0
Tor0(A H 1(C ))
Tor1(A H 1(C ))
0
0
Tor0(A H 0(C )) Tor1(A H 0(C )) 0 0
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press
8182019 Fuetterer Schwab - Spectral Sequences
httpslidepdfcomreaderfullfuetterer-schwab-spectral-sequences 1919
We now have a look at the other spectral sequence coming from the filtration by rows
From the E 0 page
P 2 otimes C 0
P 2 otimes C 1
P 2 otimes C 2
P 1 otimes C 0 P 1 otimes C 1 P 1 otimes C 2
P 0 otimes C 0 P 0 otimes C 1 P 0 otimes C 2
we as pq -th entry of the E 1 page Torq (AC p ) For q gt 0 this vanishes because every C p is
free so the E 1 page looks like this
0 0 0
A otimes C 0 A otimes C 1 A otimes C 2
Hence on the E 2 page we have again that everything except the boom row is 0 and in the
boom row we have
H 0(A otimes C lowast) H 1(A otimes C lowast) H 2(A otimes C lowast)
So by Lemma this (and hence both) spectral sequences converge to H lowast(A otimes C lowast) We
now apply Remark to the first spectral sequence For every n isin N0 we have (C)E 20n =
Tor0(A H n(C lowast)) = H n(C lowast) otimes A and (C)E 21nminus1 = Tor1(H nminus1(C lowast) A) So what we get is
the following universal coefficients theorem that describes the relation between H lowast(C lowast) and
H lowast(C lowast otimes A) in terms of the laer Tor group
Proposition ere is a short exact sequence
0 H n(C lowast) otimes A H n(C lowast otimes A) Tor1(H nminus1(C ) A) 0
References
[Wei] CA Weibel An Introduction to Homological Algebra Cambridge Studies in Ad-
vanced Mathematics Cambridge Cambridge University Press