KEVOS
ArticlesServicesCase studiesAboutContact
ArticlesServicesCase studiesAboutContact
← ArticlesRelation Matrix ConstructionEngineering · Engineering MathematicsLesson 831/884← PrevNext →
ArticlePublished 7 Aug 20262 min readBy Kevin Joginrelation matrixrelation collectionsparse matrixsmooth relations
On this page

Ask about this page

KEVOS AIRelation Matrix Construction

KEVOS knowledge first · trusted web sources when needed

Sub-exponential Class Group Computation

Relation Matrix Construction

Generating relations among ideal classes, assembling the sparse matrix, and knowing when enough relations have been collected.

Engineering / MathematicsSub-exponential Class Group Computation2 min readKV-MATH-0629

Relation collection is where the time goes. Each relation is a product of factor base ideals that turns out to be principal, recorded as an exponent vector together with its generator.

Generating candidates

Generating one relation

  1. Form a random productTake a random product of factor base ideals with small exponents.
  2. ReduceApply ideal reduction, recording the element divided out.
  3. Test smoothnessFactor the norm of the reduced ideal against the base.
  4. Record on successThe exponent vector of the original product minus that of the reduced ideal is a relation.
  5. Discard on failureMost candidates fail; this is expected.

Key point

The random product plus reduction is a random walk on the class group. Reduction brings the norm down to roughly the square root of the discriminant, which is small enough that smoothness has useful probability.

The matrix

Rows are relations, columns are factor base primes, entries are exponents. Each row also carries an associated element and its logarithmic embedding for the unit computation.

Matrix properties

SizeRows slightly exceeding the number of columns
DensityVery sparse; few non-zero entries per row
EntriesSmall integers, mostly zero, one or minus one
Side dataOne field element and one real vector per row

How many relations

Key point

Slightly more relations than factor base primes are needed — enough to make the relation lattice full rank. A surplus of a few per cent is standard, since some relations turn out to be dependent.

Relation count and correctness
ConditionConsequence
Too few relationsThe computed class number is a multiple of the truth
Just enoughCorrect, if the base generates
SurplusCorrect, with redundancy that helps verification

Improving the yield

Large prime variation

Accept relations with one prime slightly outside the base, then combine pairs sharing that prime to eliminate it. Substantially raises the yield.

Sieving

Rather than testing candidates individually, sieve over a range to identify smooth values in bulk — the same technique as in the quadratic sieve.

Better starting points

Products of small-norm primes reduce to smaller ideals, raising smoothness probability.

Parallelism

Cost

Relation collection is embarrassingly parallel: each worker searches independently and reports successes. This is where additional hardware helps, in contrast to the linear algebra stage, which does not parallelise nearly as well.

Precision in the side data

Caution

The logarithmic vector attached to each relation must be carried at sufficient precision throughout. It is accumulated across reduction steps, so error compounds, and insufficient precision corrupts the regulator without affecting the class group — producing a result that looks half-right.

Source. Henri Cohen, A Course in Computational Algebraic Number Theory, Springer GTM 138 — 6.5.2. Structural reference unverified: the source file was not available during authoring; chapter and section numbers are taken from the published edition and have not been checked against a physical copy.

Related pages

  • The Four Core Computational Tasks of Number Fields
  • Gaussian Elimination over Finite Fields
  • Recovering Abelian Group Structure from a Relation Matrix
  • Class Group and Unit Computation: the Computational Problem
  • Sub-exponential Class Group Computation for Quadratic Fields

Continue learning

Factor Base Selection and SmoothnessArticle · Engineering MathematicsNEXT LESSON →Regulator and Fundamental Unit RecoveryArticle · Engineering MathematicsIdeal Reduction in Number FieldsArticle · Engineering MathematicsVerifying Class Group and Regulator ResultsArticle · Engineering Mathematics
KEVOS · Engineering, manufacturing and project improvement
ArticlesServicesCase studiesAboutContact
© 2026 KEVOS®