The Gold Standard of Universal Algebra | Lifting a Homomorphism Through a Quotient Without Assuming a Section

Let q:B→C be a surjective homomorphism, and let g:P→C be another homomorphism. A lift of g through q is a homomorphism ℓ:P→B satisfying q∘ℓ=g. Starting at any p in P, the route through B must arrive at the same element of C as the original map g.

The domain of the unknown map is P. Its codomain is B. This simple type check prevents several plausible but unrelated calculations. We are not trying to define a map from C back to every element of B, and we are not asking whether g is constant on some unspecified quotient of P.

A section of q would be a homomorphism s:C→B with q∘s equal to the identity on C. If such a section exists, then s∘g lifts every g into C. That is a sufficient construction. It is stronger than the existence of a lift for one particular source and one particular g.

Our central examples will keep q fixed and change P. One source will impose a relation that cannot be respected upstairs; another will allow a lift. The quotient map therefore does not carry a single yes-or-no answer independent of the map being lifted.

Throughout the cyclic examples, Cn means the additive group of integers modulo n. Its generator is the residue class of 1, and n copies of that generator add to zero. Writing the modulus beside a calculation matters: an integer can represent zero in one group and a nonzero element in another.

Contents · Next chapter

Surjectivity says that for each c in C there is at least one b in B with q(b)=c. Applied to g(p), it supplies candidates for ℓ(p). It does not say that independently selected candidates will respect addition, multiplication, constants, or any other declared operation.

In an additive group, a lift must satisfy ℓ(p+p′)=ℓ(p)+ℓ(p′), as well as ℓ(0)=0. If a source generator x satisfies d x=0, its chosen image b must satisfy d b=0 upstairs. Having q(d b)=0 is weaker: it merely places d b in the kernel of q.

That difference is the obstruction we will exploit. The target quotient can hide a nonzero defect. A proposed map may appear correct after reduction while failing to preserve an equation in the algebra where the lift is supposed to live.

For a cyclic source, the whole question reduces to choosing the image of one generator and checking its defining relation. For a source with several generators, all defining relations must survive the choices. In a noncommutative group, those include multiplication relations that cannot generally be reduced to the single divisibility test developed below.

This is why a list of representatives is only the beginning of the evidence. The useful next question is not whether each representative has the right residue. It is whether different ways of expressing a source element produce the same result after the proposed substitution. A homomorphism must make those descriptions agree.

Contents · Previous chapter · Next chapter

Take q:C8→C4 to be reduction modulo four. Let the source be P=C6, and define g:C6→C4 by sending its generator to 2. This is a legitimate homomorphism because six copies of 2 give 12, which is zero modulo four.

Any lift must send the generator of C6 to an element k of C8 satisfying k≡2 modulo four. There are exactly two candidates: k=2 and k=6. We have exhausted the relevant fibre of q, so a failure for both candidates proves nonexistence.

The relation in C6 requires 6k=0 modulo eight. But 6×2=12≡4 modulo eight, and 6×6=36≡4 modulo eight. Both choices leave the same nonzero defect. Therefore no homomorphic lift exists.

Notice where the contradiction occurs. In the source, six copies of 1 equal 0, so a homomorphism must send their sum to 0. With either candidate image, the corresponding sum upstairs is 4. Reducing that defect modulo four makes it disappear, explaining why g itself was valid.

Choosing the other representative does not help. Nor does choosing representatives for all six source elements separately repair the problem: once the image of 1 is chosen, preservation of addition forces every other image. The defining relation then closes the cycle incorrectly. This is a complete obstruction, not a failure of one convenient formula that another unexplored homomorphism might avoid.

Contents · Previous chapter · Next chapter

Keep q:C8→C4, but change the source to C4. Again let g send the generator to 2. This time the source relation asks for 4k=0 modulo eight. Both candidates k=2 and k=6 satisfy it, since their fourfold sums are 8 and 24.

The two lifts are ℓ₂(x)=2x modulo eight and ℓ₆(x)=6x modulo eight. On source elements 0,1,2,3, their values are respectively 0,2,4,6 and 0,6,4,2. Reducing either list modulo four gives 0,2,0,2, exactly the values of g.

These maps preserve addition because integer multiplication distributes over addition, and changing a representative of x by four changes either output by a multiple of eight. Both algebraic preservation and independence from the chosen source representative have been checked.

Yet q has no homomorphic section. A section would have to send the generator of C4 to an element reducing to 1, hence to 1 or 5 in C8. Four copies of either candidate equal 4 modulo eight, violating the source relation. Thus this quotient has no section even though the particular map g has two lifts.

The success is tied to the map's image. Our g only reaches the subgroup {0,2} of C4. It does not demand an upstairs representative of the generator 1 that behaves like an element of order four. Requiring a lift of the identity would demand exactly that, and would fail.

Contents · Previous chapter · Next chapter

Let N=m s, so reduction modulo m gives a surjective homomorphism q:CN→Cm. Let the source be Cd, and suppose g sends its generator to the residue r modulo m. Before discussing lifts, check m divides d r; otherwise g itself is not well defined.

Every possible generator image upstairs has the form k=r+m t, where t is considered modulo s. This lists all s elements of the fibre over r without duplication. A homomorphism from Cd exists with that generator image precisely when N divides d k.

Substitute the candidate: d k=d r+d m t. Since m divides d r, define the integer a=d r/m. The divisibility condition becomes m s dividing m(a+d t), which is equivalent to a+d t being divisible by s.

The lifting problem is therefore the ordinary linear congruence d t≡−a modulo s. It has a solution exactly when gcd(d,s) divides a. When solutions exist, there are gcd(d,s) distinct solutions modulo s, and therefore that many different lifts.

Here is why the count is natural. Multiplication by d on Cs has a kernel containing gcd(d,s) elements. Every nonempty fibre of this homomorphism is a translate of that kernel, so all attainable right-hand sides have the same number of preimages. We have converted a compatibility problem into a finite equation, while keeping the validity check for g separate from the stronger lifting check.

Contents · Previous chapter · Next chapter

For the failed example, N=8, m=4, s=2, d=6 and r=2. The original map is valid because a=d r/m=3 is an integer. But gcd(6,2)=2 does not divide 3. The criterion predicts no lift, matching the explicit two-candidate calculation.

For the successful example, only d changes, from six to four. Now a=2, and gcd(4,2)=2 divides it. There are two lifts. Solving 4t≡−2 modulo two allows both residue classes t=0 and t=1, producing k=2 and k=6.

Do not insert arbitrary numbers into the gcd condition before checking m divides N and m divides d r. The first makes the stated reduction map available. The second ensures that the prescribed generator assignment defines g. A formula containing a noninteger a has already left the problem's assumptions.

Changing the integer representative of r does not alter the conclusion. Replacing r by r+m u changes a by d u. Since gcd(d,s) divides d, it divides the new a exactly when it divides the old one. The parameter t shifts to describe the same candidate elements upstairs.

The criterion is complete for these additive cyclic groups and this reduction quotient. It is not a universal test for arbitrary group presentations, lattices, or rings. Its value lies in showing precisely how a source relation becomes an arithmetic obstruction, and in giving a small setting where every possible lift can actually be counted.

Contents · Previous chapter · Next chapter

In the successful C4 example, the two lifts differ by the map δ(x)=4x modulo eight. Its values lie in ker q={0,4}. Adding this kernel-valued map to ℓ₂ produces ℓ₆, because 2x+4x=6x modulo eight.

This is not an accident restricted to those numbers. For homomorphisms of abelian groups, if ℓ and ℓ′ both lift g, their pointwise difference δ=ℓ′−ℓ is a homomorphism. Applying q gives g−g=0, so δ takes values in the kernel.

Conversely, if ℓ is one lift and δ:P→ker q is any homomorphism, then ℓ+δ is another lift. Addition preserves the homomorphism law because the groups are abelian, and q removes the added kernel contribution. Once one lift is known, all others can be described in this way.

This description separates two questions that are often blurred: whether a lift exists, and how much freedom remains after one exists. The kernel can describe the differences between successful choices without ensuring that any successful choice is available in the first place.

For nonabelian groups, pointwise products or differences of homomorphisms need not be homomorphisms. One must not carry this additive description across that boundary without additional hypotheses. In our example the arithmetic is legitimate because both source and codomain are abelian groups with addition as the operation. The setting explains why the simple subtraction argument works.

Contents · Previous chapter · Next chapter

Suppose someone proposes a table for ℓ:C4→C8 with values 0,2,0,2. Every entry reduces to the required value of g. Nevertheless it is not our valid lift ℓ₂: the source equation 1+1=2 would require ℓ(2)=ℓ(1)+ℓ(1)=4, whereas the table gives 0.

The table has copied the quotient residues into the larger carrier without preserving the operation. This is a particularly tempting error when both groups use small integer labels. Equal-looking labels do not mean the addition tables are the same.

A full finite audit can inspect all sixteen ordered input pairs in C4. For each x,y, compare the table entry at x+y modulo four with the sum of the entries at x and y modulo eight. Also check the zero and inverse operations if they are included in the declared signature.

For the formula ℓ₂(x)=2x, a proof handles all pairs at once. Write x+y=z+4u, where z is the source sum represented between zero and three. Multiplication by two gives 2x+2y=2z+8u, so both sides agree modulo eight. This also shows exactly why the wraparound is harmless.

Tables are useful evidence, but the formula proof reveals the mechanism. Use the table to catch mistakes and the relation calculation to explain them. A correct commuting equation qℓ=g alone does not establish that ℓ is a homomorphism; both requirements belong in the final certificate.

Contents · Previous chapter · Next chapter

Replace the finite cyclic source by the additive group of integers Z. A homomorphism g:Z→C4 is determined by g(1). Choose any b in C8 above that value and define ℓ(n)=n b. Then qℓ(n)=n q(b)=g(n), so this is a lift.

There is no nonzero integer d with d·1=0 in Z. The finite-order relation that obstructed the C6 example has disappeared. Negative integers cause no difficulty because a group homomorphism sends additive inverses to additive inverses.

For a free abelian group Z^r on finitely many basis vectors, choose a preimage in B of each prescribed basis image in C. Extend by integer linear combinations. If B and C are abelian groups and q is surjective, this produces a homomorphism with the required composite.

The extension is unique after the upstairs basis images have been fixed. That does not mean the lift is unique before those choices are made. Different preimages of the same target basis value can produce different lifts, just as the two candidates did in our finite example.

The word free always belongs to a class and a signature. A free group, a free abelian group and a free module have different permitted targets and relations. The basis argument here uses commuting addition in the abelian targets. Its wider universal-algebra version uses the universal property of the relevant free object, with choices made for its generators.

For a quick check of that distinction, choose the free generator image 1 in C4. Its two preimages in C8 are 1 and 5, and both define valid homomorphisms from Z. The same candidates could not define a section from C4 because fourfold addition failed there. Freedom removed a relation; it did not change the arithmetic in C8.

Contents · Previous chapter · Next chapter

A source described by generators and relations gives a practical plan. First choose upstairs preimages of the prescribed generator images. Then substitute those choices into every defining relation. Only choices satisfying all those relations can extend to a homomorphism from the presented source.

For Cd there is one generator x and the relation d x=0. Our criterion checked precisely that relation. For an abelian source with two generators and a relation 2x=3y, proposed images b and c must satisfy 2b=3c in B, not merely after applying q.

As a small illustration, let B=C8 and C=C4. Prescribe g(x)=2 and g(y)=0 for a source with relation 2x=3y. This target assignment is valid because both sides are zero modulo four. Upstairs b can be 2 or 6, while c can be 0 or 4.

Both choices for b give 2b=4 modulo eight. Taking c=0 fails, whereas taking c=4 gives 3c=12≡4 and succeeds. The generator choices are coupled by the relation. Testing their fibres independently would miss that dependency.

This example uses the abelian group presented by those generators and that relation; no additional finite-order relations are being silently imposed. If the presentation changes, the tests must change too. A complete presentation makes relation checking sufficient, while checking only some known equations establishes only a necessary condition. The distinction matters whenever a proposed lift is inferred from an incomplete list of constraints.

Contents · Previous chapter · Next chapter

An algebra P is projective relative to surjective homomorphisms in a specified class when every diagram of the stated kind can be lifted: every allowed surjection q:B→C and every homomorphism g:P→C admit a homomorphism ℓ:P→B with qℓ=g. The quantifier is the new ingredient.

One successful example does not prove that property. Our C4 source successfully lifted the map sending its generator to 2 through C8→C4. But the identity map C4→C4 does not lift through the same quotient, because such a lift would be a section, already shown impossible.

Thus C4 is not projective in the class of all abelian groups. A single failed diagram is enough to disprove the universal claim. In contrast, many successful finite trials cannot establish it unless a theorem or general construction covers every remaining diagram.

We explicitly use surjective homomorphisms here. In some algebraic categories, categorical epimorphisms need not be surjective. Replacing the stated surjections by all epimorphisms changes the requirement and should not be done merely because the words sometimes coincide in familiar examples.

This definition gives an efficient reading habit. Whenever a text calls something projective, identify the class, the morphisms and the family of maps through which lifting is required. Then distinguish a general structural guarantee from an isolated solution. Both are useful, but they answer questions with different scopes and need different evidence.

Contents · Previous chapter · Next chapter

Now restrict attention to abelian groups satisfying 4x=0 for every element. These are equivalently modules over the ring Z/4Z. In this class, C4 is free on one generator, and the obstruction used in the previous chapter is unavailable.

Given a surjection q:B→C within this restricted class and a homomorphism g:C4→C, choose b in B with q(b)=g(1). Since every element of B satisfies 4b=0, the formula ℓ(x)=x b is well defined on C4. It is a homomorphism, and qℓ=g.

That proves projectivity of C4 in this class. It does not contradict the earlier failure among all abelian groups. The counterexample used B=C8, whose generator does not satisfy 4x=0. C8 is not an allowed object in the restricted class.

This is a concrete reason to keep the ambient class visible. The underlying four-element group did not change. The set of lifting problems it was required to solve changed, and the formerly obstructing problem was excluded by an additional identity.

A useful audit is to write the class restriction next to each object in a proposed counterexample. If an upstairs algebra violates the restriction, it cannot disprove projectivity in that class. Conversely, proving a lift only for targets obeying an extra identity cannot establish projectivity in a larger class where that identity is absent. The direction of the restriction matters as much as the calculation.

Contents · Previous chapter · Next chapter

Within a variety of algebras, projectivity relative to surjections is equivalent to being a retract of a free algebra. Here a retract means there are homomorphisms i:P→F and r:F→P with r∘i equal to the identity on P. The map i embeds P, while r returns its embedded elements unchanged.

The mechanism can be seen directly. Given q:B→C and g:P→C, first compose g with r to obtain a map F→C. Because F is free, choose upstairs images for its free generators and extend them to a homomorphism h:F→B with qh=g r.

Now set ℓ=h i. Its composite with q is qh i=g r i=g. Restricting the lift from the free algebra to the retract solves the original problem. Every arrow has a role, and checking the domains makes the short composition proof readable.

For the converse, take a free algebra F with a surjective homomorphism r:F→P. If P is projective, lift its identity map through r. The resulting i:P→F satisfies r i=id, giving the required retract. This uses the existence of a free algebra mapping onto P in the variety.

For arbitrary sets of generators, the customary general argument uses choices of preimages. Our finite-generator examples require only finitely many choices. The theorem provides a structural explanation of lifting, while its hypotheses prevent us from treating every algebra that happens to sit inside a free object as projective: the retraction is essential.

Contents · Previous chapter · Next chapter

Let R=Z/6Z, and work with R-modules. Inside the free rank-one module R, consider P=3R={0,3}. The element 3 is idempotent because 3×3=9≡3 modulo six. Multiplication by 3 defines an R-linear map r:R→P.

The inclusion i:P→R is also R-linear. On the element 3, r i sends 3 to 9≡3, and it sends 0 to 0. Thus r i=id on P. This explicitly exhibits P as a retract of a free module and hence as projective.

P is not a free R-module. A free module on zero generators has one element, and a free module on any positive finite number n of generators has 6^n elements. An infinite basis gives infinitely many elements. None has exactly two elements, as P does.

We can also see the lifting construction without invoking the theorem. Given q:B→C and g:P→C, choose b with q(b)=g(3). Define ℓ(3)=3b and ℓ(0)=0. Since 2(3b)=6b=0, this respects the additive relation in P.

Moreover q(3b)=3g(3)=g(9)=g(3), so the required image is correct. Scalar compatibility follows from the same residue arithmetic: even scalars kill 3 and 3b, while odd scalars preserve them. Multiplying the arbitrary preimage by the idempotent repairs the relation. This small construction explains how projectivity can survive without a free basis of the projective object itself.

Contents · Previous chapter · Next chapter

Form the pullback E={(p,b) in P×B : g(p)=q(b)}. It is an algebra under coordinatewise operations because g and q are homomorphisms. The first-coordinate projection π:E→P is surjective: for each p, surjectivity of q supplies at least one b above g(p).

A lift ℓ defines a homomorphism s:P→E by s(p)=(p,ℓ(p)). The equality qℓ=g guarantees membership in E, and πs=id on P. Thus s is a section of this new projection.

Conversely, a section s of π must have the form s(p)=(p,ℓ(p)). Taking its second coordinate gives a homomorphism ℓ, and membership in E gives qℓ=g. A lift exists exactly when the pullback projection π has a section.

This does not reverse our conclusion that q itself may have no section. The map π has a different domain and codomain, constructed using the particular g. Confusing π with q would erase the distinction that the successful C4 example was designed to reveal.

The pullback view is useful when connecting a direct calculation to a structural lesson. It packages all eligible pairs into one object and makes compatible selection visible as a section. It does not automatically construct that section. In the failed C6 example, π remains surjective but has no homomorphic section. The obstruction has been reorganised, not removed by a new name.

Contents · Previous chapter · Next chapter

Take q:C18→C6 by reduction modulo six, and let g:C9→C6 send its generator to 2. First check the proposed map: nine copies of 2 give 18, which is zero modulo six. Its candidate generator images upstairs are 2,8 and 14.

All three candidates satisfy the source relation modulo eighteen: 9×2=18, 9×8=72 and 9×14=126 are multiples of eighteen. Hence there are three lifts. The criterion agrees: s=3, a=9×2/6=3, and gcd(9,3)=3 divides a.

Now replace C9 by C6 while retaining generator image 2. The target map is still valid, since 6×2=12 is zero modulo six. But the candidate sixfold sums are 12,48 and 84, all congruent to 12 modulo eighteen. There is no lift.

In the criterion, a becomes 2 while gcd(6,3) remains 3, which does not divide 2. The changed source relation has altered solvability despite leaving the quotient and the target generator image unchanged.

Finally retain source C6 but change its generator image to 3. The candidates are 3,9 and 15; all have sixfold sum zero modulo eighteen. This gives three lifts again. Explain both changes before relying on the formula: a lift must make the source relation vanish upstairs, and different target fibres can contain elements with different compatible orders. Correct transfer includes the reason the answer changed.

Contents · Previous chapter · Next chapter

Does a surjective quotient always admit a lift of every map? No. It gives preimages of individual values, but the C6→C4 example fails because no eligible generator image satisfies the source relation in C8. A complete obstruction checks every candidate or gives an equivalent general argument.

Is a lift necessarily injective or surjective? Neither property is required. In the successful C4→C8 example, each lift is injective but not surjective. A zero homomorphism provides other behaviour. The defining requirements are operation preservation and the commuting equation, not a claim about image size.

Does absence of a section rule out all lifts? No. It rules out a lift of the identity on the quotient. Other maps may lift, as our generator-image-two example shows. A section is a universal way to obtain lifts into that particular quotient, but an individual lift can exist without it.

If the source is projective, is the lift unique? No. Projectivity guarantees existence for the specified family of diagrams. Kernel-valued freedom can still produce several lifts. Uniqueness requires a separate argument, and the two-lift and three-lift examples make that distinction explicit.

Can I prove failure by showing that my first chosen representative fails? Only if it was the sole eligible possibility, or if an argument shows all alternatives fail for the same reason. Otherwise it proves merely that one attempt failed. In the first worked example, listing both 2 and 6 and checking both source relations made the conclusion exhaustive.

Contents · Previous chapter · Next chapter

Begin a new problem by recording the class of algebras, the declared operations and the three objects. Write q:B→C, g:P→C and the desired ℓ:P→B. Check q is surjective and g is genuinely a homomorphism before looking for an upstairs formula.

Next choose a description of P that exposes its constraints. For a cyclic source, write its generator relation. For a presented algebra, list the relations being used and explain why they are complete. For a free source, identify the actual universal property and the allowed target class.

Generate candidates for the images of the source generators, then test relations upstairs. If a candidate survives, verify that it extends to a well-defined homomorphism and that the composite equals g. If none survives, explain why the search or obstruction covered all possible homomorphisms, not just a preferred table.

After establishing existence, ask separately about uniqueness and about whether the result generalises. In abelian groups, kernel-valued differences describe alternative lifts once one is known. To claim projectivity, supply a construction for every allowed diagram or use an appropriate structural theorem with its hypotheses stated.

For independent practice, reconstruct the failed C6 example, the successful C4 example and the C18 transfer without consulting their answers. Then explain why a section of the pullback projection can exist when q has none. Being able to preserve these distinctions while changing the numbers is the practical standard: representatives, relations, lifts and sections each do a different job in one coherent calculation.

Contents · Previous chapter · Continue to related reading