Try
Consider an abelian group with two generators and three column relations. Its presentation matrix is A=[[2,4,4],[6,6,12]], so the group is Z²/AZ³. The first relation says 2e₁+6e₂=0, the second says 4e₁+6e₂=0, and the third says 4e₁+12e₂=0. Work over integers: dividing a relation by two is not a reversible move.
The matrices U=[[1,0],[3,−1]] and V=[[1,−2,−2],[0,1,0],[0,0,1]] satisfy UAV=D=[[2,0,0],[0,6,0]]. Both U and V have determinant ±1. This certificate says that the presented group is C₂⊕C₆. We will verify the certificate, explain why the diagonal factors are forced, and use the same matrices to solve equations in integers.
Predict
The matrices diag(2,6) and diag(1,12) have the same determinant. Do they present the same finite group? Compare the numbers of elements killed by multiplication by two. In C₂⊕C₆ there are four; in C₁⊕C₁₂, which is just C₁₂, there are two. The order of a finite group does not determine its structure.
Review integer presentations and Smith reduction for the column convention, reversible operations and existence of the diagonal divisibility form. The earlier finite abelian structure lesson supplies intrinsic primary decomposition and torsion-count invariants. Here we extend the result to all finitely generated abelian groups and connect it to computable certificates.
Experiment
Verify the certificate through the intermediate matrix UA=[[2,4,4],[0,6,0]]. Right multiplication by V subtracts twice the first column from each of columns two and three. The result is D. The determinants are det U=−1 and det V=1; their inverses are integer matrices.
| Quantity | Calculation from A | Calculation from D |
|---|
| Gcd of all entries | 2 | 2 |
| Two-by-two minors | −12, 0, 24 | 12, 0, 0 |
| Positive gcd of those minors | 12 | 12 |
| Rank over Q | 2 | 2 |
| Nonzero Smith factors | 2, then 12/2=6 | 2, 6 |
The individual minors changed, but the ideal they generate did not. This suggests a uniqueness invariant. Replace A by diag(1,12) and compute the first two gcds: they are 1 and 12, revealing the difference hidden by the common determinant.
Understand
Determinantal divisors prove uniqueness of the diagonal matrix
For an integer matrix A of size n by m, define Δ₀(A)=1. For k≥1, define Δₖ(A) as the nonnegative gcd of all k-by-k minors, taking it to be zero when every such minor is zero or no such minor exists. A minor is the determinant formed by choosing k rows and k columns. The rank r over Q is the largest k with a nonzero minor; the zero matrix has rank zero.
Each allowed elementary operation preserves the ideal of integers generated by the k-by-k minors. Row or column swaps and sign changes only permute or change signs of the minors. For adding an integer multiple of one row to another, expand a minor in the changed row. It is an old minor plus an integer multiple of another old minor, up to a row-order sign; if both rows were selected, the extra determinant has repeated rows and is zero. Thus every new minor is an integer combination of old minors. The inverse row operation gives the reverse containment of ideals. The column proof is the same. Therefore the positive gcd is unchanged. More generally, determinant multilinearity gives the same containment for multiplication by any integer U or V, and integer inverses give equality when they are unimodular.
Suppose D has positive nonzero diagonal entries d₁,…,dᵣ with d₁ dividing d₂, and so on, followed by zeros. Its nonzero k-minors are products of k selected diagonal entries. If the chosen indices are j₁ less than … less than jₖ, then jᵢ≥i, and dᵢ divides d indexed by jᵢ. Thus d₁…dₖ divides every such product. The minor selecting the first k diagonal entries is exactly that product. Consequently Δₖ(D)=d₁…dₖ for 1≤k≤r, and Δₖ=0 for k>r. Invariance gives
dₖ = Δₖ(A)/Δₖ₋₁(A), for 1≤k≤r.
The denominator is positive in this range. Do not apply the ratio formula past the rank, where a zero denominator may appear. It follows that the ordered nonzero Smith factors of a fixed matrix are unique. Changing the reduction path cannot change them. This proves matrix uniqueness; abstract group uniqueness across different-sized presentations still needs its own argument below.
Classify the quotient, including free directions
Write C(d)=Z/dZ for the cyclic group of order d. For D of size n by m and rank r, its image consists of vectors whose first r coordinates are multiples of d₁,…,dᵣ and whose remaining n−r coordinates are zero. Reducing these first coordinates modulo their diagonal entries gives a surjective map
Zⁿ → C(d₁) ⊕ … ⊕ C(dᵣ) ⊕ Z^(n−r).
Its kernel is exactly DZᵐ. The first isomorphism theorem therefore identifies this direct sum with the presented group. A factor dᵢ=1 contributes the trivial group and can be omitted. Zero rows give free infinite cyclic factors; zero columns represent redundant relation coordinates and do not themselves add group generators.
Every finitely generated abelian group has a finite presentation: a chosen generating list gives a surjection from Zⁿ, and the preceding lesson proved that its kernel is finitely generated. Apply Smith reduction to a finite matrix of those kernel generators. Thus every such group is isomorphic to Zᶠ⊕C(d₁)⊕…⊕C(dₜ) with d₁>1 and d₁ dividing … dividing dₜ. Here f≥0 and t≥0; the trivial group has both zero. This uses a theorem about all subgroups of Zⁿ, not a claim that a convenient relation matrix is always immediately visible.
Why different presentations give the same abstract invariants
The torsion subgroup T consists of elements killed by some positive integer. In an abelian group it is a subgroup: if ax=0 and by=0, then ab(x−y)=0. An isomorphism preserves T. In the displayed decomposition T is the finite cyclic-factor part, and G/T is Zᶠ. The quotient (G/T)/2(G/T) has exactly 2ᶠ elements. Its cardinality is intrinsic, so it determines f without referring to a presentation.
For each prime p, the subgroup Tₚ consists of torsion elements killed by some power of p; it too is intrinsic. By the finite primary decomposition, T is the direct sum of these Tₚ. If Tₚ is a sum of cyclic factors of orders p raised to positive exponents aᵢ, the quotient p^(k−1)Tₚ / pᵏTₚ is a vector space over Fₚ. Each cyclic factor contributes dimension one exactly when aᵢ≥k, and contributes zero otherwise. Therefore its dimension cₖ counts the exponents at least k. The difference cₖ−cₖ₊₁ counts exponents equal to k, recovering the full multiset for each p.
To recover the invariant factors, sort each prime’s positive exponent list in increasing order and pad it on the left with zeros to the largest list length t. Multiply the corresponding prime powers position by position. The resulting d₁,…,dₜ obey divisibility. This alignment is forced by the divisibility chain, and at least one prime’s unpadded list has length t, so every dᵢ>1. The finite cyclic CRT justifies combining relatively prime factors at each position. Hence the nontrivial invariant factors and free rank are unique as group invariants, even if two presentations have different numbers of generators, relations or unit factors.
For example C₄⊕C₆ has 2-primary exponents (1,2) and 3-primary exponent (1). Pad the latter to (0,1); the invariant factors are 2 and 12. Thus C₄⊕C₆≅C₂⊕C₁₂. This is not a license to replace any pair of cyclic orders by their product: the shared prime parts must be accounted for.
Solve integer equations with the same certificate
Let A have n rows and m columns, and let UAV=D. To solve Ax=b for x∈Zᵐ, put x=Vz and c=Ub. Since U and V have integer inverses, Ax=b is equivalent to Dz=c. There are solutions precisely when dᵢ divides cᵢ for every 1≤i≤r and cᵢ=0 for every row i>r. When consistent, zᵢ=cᵢ/dᵢ for i≤r and all remaining m−r coordinates of z are arbitrary integers. Multiplication by V gives every solution x, without omissions or duplicates. The zero-column directions give the kernel basis consisting of columns r+1,…,m of V.
In our worked example c=(b₁,3b₁−b₂). Consistency requires b₁ even and 6 dividing 3b₁−b₂. For b=(2,0), the transformed solutions are z=(1,1,t), t∈Z. Therefore every solution is x=(−1−2t,1,t). Check directly: the first row gives −2−4t+4+4t=2, and the second gives −6−12t+6+12t=0. For b=(2,1), the second transformed coordinate is five, which is not divisible by six; there is no integer solution even though a rational solution exists.
The quotient isomorphism in the same example sends the class of b to (b₁ modulo 2, 3b₁−b₂ modulo 6). This connects the abstract group obstruction with the arithmetic consistency test. Group cardinality, exact solutions and integer kernels are three uses of one verified certificate.
Apply
Change to the rectangular matrix B=[[2,0],[0,6],[0,0]]. Classify Z³/BZ², then solve Bx=(4,12,q) over integers as q varies. Which coordinates create a free quotient direction, and which create a free solution parameter? Finally compare B with diag(1,2,6,0), viewed as a four-by-four presentation. Explain why different matrix sizes need not mean different groups.
Check zero rows, unit factors and solution parameters
B is already in Smith form with rank two. Its quotient is C₂⊕C₆⊕Z: the third zero row gives the free quotient direction. The equation is consistent exactly when q=0, and then x=(2,2) is unique because there are no extra domain coordinates; the zero row does not create a solution parameter. The four-by-four diagonal matrix has rank three and factors 1,2,6. Its quotient is C₁⊕C₂⊕C₆⊕Z, isomorphic to the same group after omitting the unit factor. Its extra zero column does create a kernel direction for its own equation system. Classifying the quotient and solving the matrix equation are related but different questions.
Continue
For a new presentation, supply the matrices U,V and multiply them out; then compute determinantal divisors as an independent check. Report free rank, nontrivial cyclic factors, integer consistency conditions and a complete solution family where requested. The saved tasks and mixed rings-and-fields assessment ask for proofs as well as calculations. Extending this machinery to polynomial rings and rational canonical form remains a separate next step; the operator-module definition alone does not prove that theory.
