Documentation

Complexitylib.Algebraic.LowerBound.Fusion.SumOfTerms.Waring.Rectangular

Rectangular-degree catalecticants for Waring sums #

Parameterize the squarefree Waring flattening by an arbitrary total degree d and layer split k. Both matrix axes are indexed by k-subsets; the column is complemented before forming the queried exponent, so its contribution has degree d-k. The squarefree target becomes a scalar identity matrix of dimension choose d k, while every d-th power of a linear form remains rank one.

One scalar multiple of a degree-d power of a linear form.

  • scale : K

    Scalar multiplying the power.

  • coefficients : Fin degree → K

    Coefficients of the linear form in degree variables.

Instances For
    noncomputable def Algebraic.Fusion.SumOfTerms.Waring.Rectangular.linearForm {K : Type} [CommSemiring K] {degree : ℕ} (term : Term K degree) :
    MvPolynomial (Fin degree) K

    Linear form represented by a rectangular-degree Waring term.

    Equations
    Instances For
      noncomputable def Algebraic.Fusion.SumOfTerms.Waring.Rectangular.termValue {K : Type} [CommSemiring K] {degree : ℕ} (term : Term K degree) :
      MvPolynomial (Fin degree) K

      Polynomial value of a charged degree-d Waring term.

      Equations
      Instances For
        noncomputable def Algebraic.Fusion.SumOfTerms.Waring.Rectangular.complementSet {degree split : ℕ} (set : MatrixRank.Layer degree split) :
        Finset (Fin degree)

        Complement of a layer index, viewed as a set rather than forced back into the same layer.

        Equations
        Instances For
          @[simp]
          noncomputable def Algebraic.Fusion.SumOfTerms.Waring.Rectangular.entryExponent (degree split : ℕ) (row column : MatrixRank.Layer degree split) :
          Fin degree →₀ ℕ

          Exponent queried by the split-k catalecticant entry.

          Equations
          • One or more equations did not get rendered due to their size.
          Instances For
            theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.entryExponent_eq_targetExponent_iff (degree split : ℕ) (row column : MatrixRank.Layer degree split) :
            entryExponent degree split row column = exponent Finset.univ ↔ row = column

            A row set plus a complemented column set is the all-ones exponent exactly on the matrix diagonal.

            theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.entryExponent_sum (degree split : ℕ) (row column : MatrixRank.Layer degree split) :
            ((entryExponent degree split row column).sum fun (x : Fin degree) (multiplicity : ℕ) => multiplicity) = degree

            Every queried entry exponent has total degree degree.

            noncomputable def Algebraic.Fusion.SumOfTerms.Waring.Rectangular.catalecticant (K : Type) [Field K] (degree split : ℕ) :
            MvPolynomial (Fin degree) K →ₗ[K] Matrix (MatrixRank.Layer degree split) (MatrixRank.Layer degree split) K

            Normalized degree-d, split-k catalecticant.

            Equations
            • One or more equations did not get rendered due to their size.
            Instances For
              @[simp]
              theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.catalecticant_apply {K : Type} [Field K] (degree split : ℕ) (polynomial : MvPolynomial (Fin degree) K) (row column : MatrixRank.Layer degree split) :
              (catalecticant K degree split) polynomial row column = polynomial.coeff (entryExponent degree split row column) / ↑(entryExponent degree split row column).multinomial
              theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.entryExponent_multinomial_ne_zero {K : Type} [Field K] [CharZero K] (degree split : ℕ) (row column : MatrixRank.Layer degree split) :
              ↑(entryExponent degree split row column).multinomial ≠ 0

              Every normalization denominator is nonzero in characteristic zero.

              theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.entryExponent_prod {degree split : ℕ} {K : Type} [CommMonoid K] (coefficients : Fin degree → K) (row column : MatrixRank.Layer degree split) :
              ((entryExponent degree split row column).prod fun (index : Fin degree) (multiplicity : ℕ) => coefficients index ^ multiplicity) = (∏ index ∈ ↑row, coefficients index) * ∏ index ∈ complementSet column, coefficients index

              Coefficient products factor across the row and complemented column.

              theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.coeff_linearForm_pow_entryExponent {degree split : ℕ} {K : Type} [CommSemiring K] (term : Term K degree) (row column : MatrixRank.Layer degree split) :
              (linearForm term ^ degree).coeff (entryExponent degree split row column) = ↑(entryExponent degree split row column).multinomial * ((∏ index ∈ ↑row, term.coefficients index) * ∏ index ∈ complementSet column, term.coefficients index)

              Coefficient formula for a linear-form power at a queried exponent.

              def Algebraic.Fusion.SumOfTerms.Waring.Rectangular.leftVector {degree split : ℕ} {K : Type} [CommMonoid K] (term : Term K degree) :
              MatrixRank.Layer degree split → K

              Row vector in the rank-one rectangular catalecticant of a power term.

              Equations
              Instances For
                noncomputable def Algebraic.Fusion.SumOfTerms.Waring.Rectangular.rightVector {degree split : ℕ} {K : Type} [CommMonoid K] (term : Term K degree) :
                MatrixRank.Layer degree split → K

                Complement-reindexed column vector.

                Equations
                Instances For
                  theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.catalecticant_termValue {degree : ℕ} {K : Type} [Field K] [CharZero K] (term : Term K degree) (split : ℕ) :
                  (catalecticant K degree split) (termValue term) = Matrix.vecMulVec (leftVector term) (rightVector term)

                  A normalized rectangular catalecticant of a power term is an outer product.

                  theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.target_eq_prod_X (K : Type) [CommSemiring K] (degree : ℕ) :
                  target K degree = ∏ index : Fin degree, MvPolynomial.X index

                  The generalized target is literally the product of all variables.

                  The inverse, in K, of the multinomial coefficient of the target exponent; it appears on the target catalecticant diagonal. It is nonzero in characteristic zero (targetScalar_ne_zero).

                  Equations
                  Instances For

                    The target normalizing scalar is nonzero in characteristic zero.

                    theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.catalecticant_target {K : Type} [Field K] (degree split : ℕ) :
                    (catalecticant K degree split) (target K degree) = targetScalar K degree • 1

                    The squarefree target has a scalar identity at every layer split.

                    noncomputable def Algebraic.Fusion.SumOfTerms.Waring.Rectangular.feature (K : Type) [Field K] (degree split : ℕ) :
                    MvPolynomial (Fin degree) K →ₗ[K] (MatrixRank.Layer degree split → K) →ₗ[K] MatrixRank.Layer degree split → K

                    Rectangular catalecticant followed by matrix-to-linear-map conversion.

                    Equations
                    Instances For
                      theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.feature_target {K : Type} [Field K] (degree split : ℕ) :
                      (feature K degree split) (target K degree) = targetScalar K degree • LinearMap.id

                      The target feature is targetScalar K degree times the identity map, over any field. The scalar is nonzero in characteristic zero (targetScalar_ne_zero).

                      theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.target_rank_ge {K : Type} [Field K] [CharZero K] (degree split : ℕ) :
                      ↑(degree.choose split) ≤ ((feature K degree split) (target K degree)).rank

                      Target rank is the full choose degree split layer dimension.

                      theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.targetRank_le_middle (degree split : ℕ) :
                      degree.choose split ≤ degree.choose (degree / 2)

                      Among the rectangular splits, the middle layer maximizes the raw target rank. Off-center splits are useful only when they improve the corresponding local interaction-rank bound.

                      theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.term_rank_le_one {degree : ℕ} {K : Type} [Field K] [CharZero K] (term : Term K degree) (split : ℕ) :
                      ((feature K degree split) (termValue term)).rank ≤ 1

                      Every charged degree-d power has rectangular feature rank at most one.

                      @[reducible, inline]

                      Construct the degree-d squarefree monomial from degree-d powers.

                      Equations
                      • One or more equations did not get rendered due to their size.
                      Instances For

                        Full rectangular-rank certificate for the squarefree target.

                        Equations
                        • One or more equations did not get rendered due to their size.
                        Instances For
                          @[simp]
                          theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.certificate_targetRank (K : Type) [Field K] [CharZero K] (degree split : ℕ) :
                          (certificate K degree split).targetRank = degree.choose split
                          theorem Algebraic.Fusion.SumOfTerms.Waring.Rectangular.choose_lowerBound {K : Type} [Field K] [CharZero K] (degree split : ℕ) (circuit : Circuit (SumOfTerms.signature (Term K degree)) 0 1) (constructs : (problem K degree).Constructs circuit (SumOfTerms.interpretation termValue)) :
                          degree.choose split ≤ circuit.cost SumOfTerms.termCost

                          Split-k rectangular catalecticants force choose d k Waring terms.

                          The middle split recovers the central-binomial lower bound at even degree.