nLab
Tor

Context

Homological algebra

homological algebra

and

nonabelian homological algebra

Context

Basic definitions

Stable homotopy theory notions

Constructions

Lemmas

diagram chasing

Homology theories

Theorems

Contents

Idea

In the context of homological algebra, the Tor-functor is the
left derived functor of the tensor product of R-modules, for R a commutative ring.

Together with the Ext-functor it constitutes one of the central operations of interest in homological algebra.

Definition

Given a ring R the bifunctor R:Mod R× RModAb from two copies of R-Mod to Ab is a right exact functor. Its left derived functors are the Tor-functors

Tor(,B):Mod RAbTor(-,B) : Mod_R \to Ab

and

Tor(A,):Mod RAbTor(A,-) : Mod_R \to Ab

with respect to one argument with fixed another, if they exist, are parts of a bifunctor

Tor:Mod R× RModAb.Tor : Mod_{R}\times _{R}Mod\to Ab \,.

Properties

Existence and balancing

Given a right R-module

AMod RA \in Mod_R

and a left R-module

B RModB \in {}_R Mod

there are in principle three different ways to compute their derived tensor product Tor (A,B):

  1. keeping B fixed and deriving the functor

    () RB:Mod RAb(-) \otimes_R B : Mod_R \to Ab
  2. keeping A fixed and deriving the functor

    A R(): RModAbA \otimes_R (-) : {}_R Mod \to Ab
  3. deriving the functor

    () R():Mor R× RModAb(-) \otimes_R (-) : Mor_R \times {}_R Mod \to Ab

    in both arguments

Theorem

If both Mod R and RMod have enough projectives, then all these three derived functors exist and all give the same result.

Proof

Existence is clear from the very definition of derived functor in homological algebra. So we show that deriving in the left argument gives the same result as deriving in the right argument.

Let Q A qiA and Q B qiB be projective resolutions of A and B, respectively. The corresponding tensor product of chain complexes]] Tot(Q AQ B), hence by prop. \ref{AsTotalComplex} the total complex of the degreewise tensor product of modules double complex carries the filtration by horizontal degree as well as that by vertical degree.

Accordingly there are the corresponding two spectral sequences of a double complex, to be denoted here { AE p,q r} r,p,q (for the filtering by A-degree) and { BE p,q r} r,p,q (for the filtering by B-degree). By the discussion there, both converge to the chain homology of the total complex.

We find the value of both spectral sequences on low degree pages according to the general discussion at spectral sequence of a double complex - low degree pages.

The 0th page for both is

AE p,q 0= BE p,q 0Q p A RQ q B.{}^A E^0_{p,q} = {}^B E^0_{p,q} \coloneqq Q^A_p \otimes_R Q^B_q \,.

For the first page we have

AE p,q 1 H q(C p,) H q(Q p AQ B)\begin{aligned} {}^A E^1_{p,q} & \simeq H_q(C_{p,\bullet}) \\ & \simeq H_q( Q^A_p \otimes Q^B_\bullet ) \end{aligned}

and

BE p,q 1 H q(C ,p) H q(Q AQ p B).\begin{aligned} {}^B E^1_{p,q} & \simeq H_q(C_{\bullet,p}) \\ & \simeq H_q( Q^A_\bullet \otimes Q^B_p ) \end{aligned} \,.

Now using the universal coefficient theorem in homology and the fact that Q A and Q B is a resolution by projective objects, by construction, hence of tensor acyclic objects for which all Tor-modules vanish, this simplifies to

AE p,q 1 Q p AH q(Q B) {Q p A RB ifq=0 0 otherwise\begin{aligned} {}^A E^1_{p,q} & \simeq Q^A_p \otimes H_q(Q^B_\bullet) \\ & \simeq \left\{ \array{ Q^A_p \otimes_R B & if\; q = 0 \\ 0 & otherwise } \right. \end{aligned}

and similarly

BE p,q 1 H q(Q A) RQ p B {A RQ p B ifq=0 0 otherwise.\begin{aligned} {}^B E^1_{p,q} & \simeq H_q(Q^A_\bullet) \otimes_R Q^B_p \\ & \simeq \left\{ \array{ A \otimes_R Q^B_p & if\; q = 0 \\ 0 & otherwise } \right. \end{aligned} \,.

It follows for the second pages that

AE p,q 2 H p(H q vert(Q AQ B)) {(L p(() RB))(A) ifq=0 0 otherwise\begin{aligned} {}^A E^2_{p,q} & \simeq H_p(H^{vert}_q(Q^A_\bullet \otimes Q^B_\bullet)) \\ & \simeq \left\{ \array{ (L_p( (-)\otimes_R B ))(A) & if \; q = 0 \\ 0 & otherwise } \right. \end{aligned}

and

BE p,q 2 H p(H q hor(Q AQ B)) {(L p(A R()))(B) ifq=0 0otherwise.\begin{aligned} {}^B E^2_{p,q} & \simeq H_p(H^{hor}_q(Q^A_\bullet \otimes Q^B_\bullet)) \\ & \simeq \left\{ \array{ (L_p ( A \otimes_R (-) ))(B) & if \; q = 0 \\ 0 \; otherwise } \right. \end{aligned} \,.

Now both of these second pages are concentrated in a single row and hence have converged on that page already. Therefore, since they both converge to the same value:

L p(() RB)(A) AE p,0 2 AE p,0 BE p,0 2L p(A R())(B).L_p((-)\otimes_R B)(A) \simeq {}^A E^2_{p,0} \simeq {}^A E^\infty_{p,0} \simeq {}^B E^2_{p,0} \simeq L_p(A \otimes_R (-))(B) \,.

Respect for direct sums and filtered colimits

Proposition

Each Tor n R(,N) respects direct sums.

Proof

Let S Set and let {N s} sS be an S-family of R-modules. Observe that

  1. if {(F s) } sS is an family of projective resolutions, then their degreewise direct sum ( sSF) is a projective resolution of sSN s.

  2. the tensor product functor distributes over direct sums (this is discussed at tensor product of modules – monoidal category structure)

  3. the chain homology functor preserves direct sums (this is discussed at chain homology - respect for direct sums).

Using this we have

Tor n R( sSN s,N) H n(( sSF)N) H n( sS(F sN)) sSH n(F sN) sSTor n(N s,N).\begin{aligned} Tor_n^R(\oplus_{s \in S} N_s, N) & \simeq H_n\left( \left(\oplus_{s \in S} F\right) \otimes N \right) \\ & \simeq H_n\left( \oplus_{s \in S} \left(F_s \otimes N \right) \right) \\ & \simeq \oplus_{s \in S} H_n( F_s \otimes N ) \\ & \simeq \oplus_{s \in S} Tor_n(N_s, N) \end{aligned} \,.
Proposition

Each Tor n R(,N) respects filtered colimits.

Proof

Let hence A:IRMod be a filtered diagram of modules. For each A i, iI we may find a projective resolution and in fact a free resolution (Y i) qiA. Since chain homology commutes with filtered colimits (this is discussed at chain homology - respect for filtered colimits), this means that

(lim iY i) A(\underset{\to_i}{\lim} Y_i)_\bullet \to A

is still a quasi-isomorphism. Moreover, by Lazard's criterion the degreewise filtered colimits of free modules lim i(Y i) n for each n are flat modules. This means that lim i(Y i) A is flat resolution of A. By the very definition or else by the basic properties of flat modules, this means that it is a ()N-acyclic resolution. By the discussion there it follows that

Tor n (A,N)H n((lim iY i)N).Tor_n^\mathbb{Z}(A,N) \simeq H_n( (\underset{\to_i}{\lim} Y_i) \otimes N ) \,.

Now the tensor product of modules is a left adjoint functor (the right adjoint being the internal hom of modules) and so it commutes over the filtered colimit to yield, using again that chain homology commutes with filtered colimits,

H n(lim i(Y iN)) lim iH n(Y iN) lim iTor n(A i,N).\begin{aligned} \cdots & \simeq H_n( \underset{\to_i}{\lim} (Y_i \otimes N) ) \\ & \simeq \underset{\to_i}{\lim} H_n( Y_i \otimes N ) \\ & \simeq \underset{\to_i}{\lim} Tor_n( A_i, N) \end{aligned} \,.

Relation to torsion groups

An abelian group is called torsion if its elements are “nilpotent”, hence if all its elements have finite order.

Definition

For A Ab and p, write

pA{aApa=0}{}_p A \coloneqq \{ a \in A | p \cdot a = 0 \}

for the p-torsion subgroup consisting of all those elements whose p-fold sum with themselves gives 0.

For n with n1, write n=/n for the cyclic group of order n, as usual.

Proposition

For p, p1, and A Ab Mod any abelian group, we have an isomorphism

Tor 1 ( p,A) pATor_1^\mathbb{Z}(\mathbb{Z}_p, A) \simeq {}_p A

of the Tor 1-group with the p-torsion subgroup of A.

For p=0 we have

Tor 1 (,A)0.Tor_1^{\mathbb{Z}}(\mathbb{Z}, A) \simeq 0 \,.
Proof

For the first statement, the short exact sequence

0pmodp p00 \to \mathbb{Z} \stackrel{\cdot p}{\to} \mathbb{Z} \stackrel{mod\, p}{\to} \mathbb{Z}_p \to 0

constitutes a projective resolution (even a free resolution) of p. Accordingly we have

Tor 1 ( p,A) H 1([0A(p)AA) ker((p)A) {aApa=0}.\begin{aligned} Tor_1^\mathbb{Z}(\mathbb{Z}_p, A) &\simeq H_1( [\cdots\to 0 \to \mathbb{Z}\otimes A \stackrel{(\cdot p) \otimes A}{\to} \mathbb{Z} \otimes A ) \\ & \simeq ker( (\cdot p) \otimes A ) \\ & \simeq \{ a\in A | p\cdot a = 0 \} \end{aligned} \,.

Here in the last step we use that (p)A acts as

(1,a) (p,a) =p(1,a) =(1,pa).\begin{aligned} (1, a) &\mapsto (p,a) \\ & = p \cdot (1,a) \\ & = (1, p \cdot a) \end{aligned} \,.

For the second statement, is already free hence [00] is already a projective resolution and hence Tor 1(,A)H 1(0)0.

Proposition

Let A be a finite abelian group and B any abelian group. Then Tor 1(A,B) is a torsion group. Specifically, Tor 1(A,B) is a direct sum of torsion subgroups of A.

Proof

By a fundamental fact about finite abelian groups (see this theorem), A is a direct sum of cyclic group Aoplusk p k. By prop. 1 Tor 1 respects this direct sum, so that

Tor 1(A,B) kTor 1( p k,B).Tor_1(A,B) \simeq \oplus_k Tor_1(\mathbb{Z}_{p_k}, B) \,.

By prop. 3 every direct summand on the right is a torsion group and hence so is the whole direct sum.

More generally we have:

Proposition

Let A and B be abelian groups. Write Tor for the left derived functor of tensoring over R=. Then

  1. Tor 1 (A,B) is a torsion group. Specifically it is a filtered colimit of torsion subgroups of B.

  2. Tor 1 (/,A) is the torsion subgroup of A.

  3. A is a torsion-free group precisely if Tor 1 (A,)=0, equivalently if Tor 1 (,A)=0.

For instance (Weibel, prop. 3.1.2, prop. 3.1.3, cor. 3.1.5).

Proof

The group A may be expressed as a filtered colimit

Alim iA iA \simeq \underset{\to_i}{\lim} A_i

of finitely generated subgroups (this is discussed at Mod - Limits and colimits). Each of these is a direct sum of cyclic groups.

By prop. 2 Tor 1 (,B) preserves these colimits. By prop. 3 every cyclic group is sent to a torsion group (of either A or B).Thereforebyprop.refTorOutOfCyclicGroupTor_1(A,B)$ is a filtered colimit of direct sums of torsion groups. This is itself a torsion group.

Remark

Analogous results fail, in general, for replaced by another ring R.

Corollary

An abelian group is torsion free precisely if regarded as a -module it is a flat module.

See at flat module - Examples for more.

Symmetry in the two arguments

Proposition

For N 1,N 2RMod and n there is a natural isomorphism

Tor n(A,B)Tor n(B,A).Tor_n(A,B) \simeq Tor_n(B,A) \,.

We first give a proof for R a principal ideal domain such as .

Proof

Let R be a principal ideal domain such as (in the latter case RMod Ab). Then by the discussion at projective resolution – length-1 resolutions there is always a short exact sequence

0F 1F 0N00 \to F_1 \to F_0 \to N \to 0

exhibiting a projective resolution of any module N. It follows that Tor n2(,)=0.

Let then 0F 1F 2N 20 be such a short resolution for N 2. Then by the long exact sequence of a derived functor this induces an exact sequence of the form

0Tor 1(N 1,F 1)Tor 1(N 1,F 0)Tor 1(N 1,N 2)N 1F 1N 1F 0N 1N 20.0 \to Tor_1(N_1, F_1) \to Tor_1(N_1, F_0) \to Tor_1(N_1, N_2) \to N_1 \otimes F_1 \to N_1 \otimes F_0 \to N_1 \otimes N_2 \to 0 \,.

Since by construction F 0 and F 1 are already projective modules themselves this collapses to an exact sequence

0Tor 1(N 1,N 2)N 1F 1N 1F 0N 1N 20.0 \to Tor_1(N_1, N_2) \hookrightarrow N_1 \otimes F_1 \to N_1 \otimes F_0 \to N_1 \otimes N_2 \to 0 \,.

To the last three terms we apply the natural symmetric braiding isomorphism in (RMod, R) to get

0 Tor 1(N 1,N 2) N 1F 1 N 1F 0 N 1N 2 0 0 Tor 1(N 2,N 1) F 1N 1 F 0N 1 N 2N 1 0.\array{ 0 &\to& Tor_1(N_1, N_2) &\hookrightarrow& N_1 \otimes F_1 &\to& N_1 \otimes F_0 &\to& N_1 \otimes N_2 &\to& 0 \\ && \downarrow && \downarrow^{\mathrlap{\simeq}} && \downarrow^{\mathrlap{\simeq}} && \downarrow^{\mathrlap{\simeq}} && \\ 0 &\to& Tor_1(N_2, N_1) &\hookrightarrow& F_1 \otimes N_1 &\to& F_0 \otimes N_1 &\to& N_2 \otimes N_1 &\to& 0 } \,.

This exhibits a morphism Tor 1(N 1,N 2)Tor 1(N 2,N 1) as the morphism induced on kernels from an isomorphism between two morphisms. Hence this is itself an isomorphism. (This is just by the universal property of the kernel, but one may also think of it as a simple application of the the four lemma/five lemma.)

Localization

(…)

For instance (Weibel, cor. 3.2.13).

References

Standard textbook accounts include the following:

Lecture notes include

  • Daniel Murfet, Tor (pdf)

section 3 of

Original articles include

  • Patrick Keef, On the Tor functor and some classes of abelian groups, Pacific J. Math. Volume 132, Number 1 (1988), 63-84. (Euclid)

Revised on October 29, 2012 17:37:59 by Urs Schreiber (131.174.188.167)