Description of fast matrix multiplication algorithm: ⟨4×5×6:93⟩

Algorithm type

X4Y4Z4+2X3Y3Z4+2X2Y4Z2+2X2Y3Z2+16X2Y2Z2+2XY3Z2+2X2YZ2+18XY2Z+48XYZX4Y4Z42X3Y3Z42X2Y4Z22X2Y3Z216X2Y2Z22XY3Z22X2YZ218XY2Z48XYZX^4*Y^4*Z^4+2*X^3*Y^3*Z^4+2*X^2*Y^4*Z^2+2*X^2*Y^3*Z^2+16*X^2*Y^2*Z^2+2*X*Y^3*Z^2+2*X^2*Y*Z^2+18*X*Y^2*Z+48*X*Y*Z

Algorithm definition

The algorithm ⟨4×5×6:93⟩ could be constructed using the following decomposition:

⟨4×5×6:93⟩ = ⟨2×2×3:11⟩ + ⟨2×2×3:11⟩ + ⟨2×3×3:15⟩ + ⟨2×3×3:15⟩ + ⟨2×3×3:15⟩ + ⟨2×2×3:11⟩ + ⟨2×3×3:15⟩.

This decomposition is defined by the following equality:

TraceMulA_1_1A_1_2A_1_3A_1_4A_1_5A_2_1A_2_2A_2_3A_2_4A_2_5A_3_1A_3_2A_3_3A_3_4A_3_5A_4_1A_4_2A_4_3A_4_4A_4_5B_1_1B_1_2B_1_3B_1_4B_1_5B_1_6B_2_1B_2_2B_2_3B_2_4B_2_5B_2_6B_3_1B_3_2B_3_3B_3_4B_3_5B_3_6B_4_1B_4_2B_4_3B_4_4B_4_5B_4_6B_5_1B_5_2B_5_3B_5_4B_5_5B_5_6C_1_1C_1_2C_1_3C_1_4C_2_1C_2_2C_2_3C_2_4C_3_1C_3_2C_3_3C_3_4C_4_1C_4_2C_4_3C_4_4C_5_1C_5_2C_5_3C_5_4C_6_1C_6_2C_6_3C_6_4=TraceMulA_3_3-A_3_4A_3_2-A_3_5A_4_3-A_4_4A_4_2-A_4_5-B_3_2+B_3_1-B_3_5+B_3_4-B_3_6+B_3_3B_2_1-B_2_2B_2_4-B_2_5B_2_3-B_2_6C_1_3-C_1_2-C_1_1+C_1_4-C_4_2+C_4_3-C_4_1+C_4_4-C_3_2+C_3_3-C_3_1+C_3_4+TraceMulA_3_3+A_2_3A_2_2+A_3_2A_1_3+A_4_3A_1_2+A_4_2B_3_1+B_4_1B_4_4+B_3_4B_3_3+B_4_3B_2_1+B_5_1B_2_4+B_5_4B_2_3+B_5_3C_1_3+C_2_3C_1_4+C_2_4C_4_3+C_5_3C_5_4+C_4_4C_3_3+C_6_3C_3_4+C_6_4+TraceMulA_2_1A_2_4A_2_5A_1_1A_1_4A_1_5B_1_2B_1_5B_1_6B_4_2B_4_5B_4_6B_5_2B_5_5B_5_6C_2_2C_2_1C_5_2C_5_1C_6_2C_6_1+TraceMulA_3_1A_3_4A_3_5A_4_1A_4_4A_4_5-B_1_1+B_1_2-B_1_4+B_1_5-B_1_3+B_1_6-B_3_1-B_4_1+B_3_2+B_4_2-B_4_4-B_3_4+B_3_5+B_4_5-B_3_3-B_4_3+B_4_6+B_3_6-B_2_1-B_5_1+B_5_2+B_2_2-B_5_4-B_2_4+B_2_5+B_5_5-B_2_3-B_5_3+B_2_6+B_5_6C_2_3C_2_4C_5_3C_5_4C_6_3C_6_4+TraceMulA_3_1+A_2_1-A_2_3-A_3_3+A_2_4+A_3_4-A_3_2-A_2_2+A_2_5+A_3_5A_1_1+A_4_1-A_1_3-A_4_3+A_1_4+A_4_4-A_1_2-A_4_2+A_1_5+A_4_5B_1_1B_1_4B_1_3B_4_1B_4_4B_4_3B_5_1B_5_4B_5_3C_1_2C_1_1C_4_2C_4_1C_3_2C_3_1+TraceMulA_2_3A_2_2A_1_3A_1_2B_3_2B_3_5B_3_6B_2_2B_2_5B_2_6-C_1_3-C_2_3+C_2_2+C_1_2C_1_1+C_2_1-C_1_4-C_2_4C_4_2+C_5_2-C_4_3-C_5_3C_4_1+C_5_1-C_4_4-C_5_4C_3_2+C_6_2-C_3_3-C_6_3C_3_1+C_6_1-C_3_4-C_6_4+TraceMulA_3_1-A_2_3-A_3_3+A_3_4-A_3_2-A_2_2+A_3_5A_4_1-A_1_3-A_4_3+A_4_4-A_1_2-A_4_2+A_4_5B_1_1B_1_4B_1_3B_3_1+B_4_1-B_3_2B_4_4+B_3_4-B_3_5B_3_3+B_4_3-B_3_6B_2_1+B_5_1-B_2_2B_2_4+B_5_4-B_2_5B_2_3+B_5_3-B_2_6C_1_3-C_1_2+C_2_3-C_1_1+C_1_4+C_2_4-C_4_2+C_4_3+C_5_3-C_4_1+C_4_4+C_5_4-C_3_2+C_3_3+C_6_3-C_3_1+C_3_4+C_6_4TraceMulA_1_1A_1_2A_1_3A_1_4A_1_5A_2_1A_2_2A_2_3A_2_4A_2_5A_3_1A_3_2A_3_3A_3_4A_3_5A_4_1A_4_2A_4_3A_4_4A_4_5B_1_1B_1_2B_1_3B_1_4B_1_5B_1_6B_2_1B_2_2B_2_3B_2_4B_2_5B_2_6B_3_1B_3_2B_3_3B_3_4B_3_5B_3_6B_4_1B_4_2B_4_3B_4_4B_4_5B_4_6B_5_1B_5_2B_5_3B_5_4B_5_5B_5_6C_1_1C_1_2C_1_3C_1_4C_2_1C_2_2C_2_3C_2_4C_3_1C_3_2C_3_3C_3_4C_4_1C_4_2C_4_3C_4_4C_5_1C_5_2C_5_3C_5_4C_6_1C_6_2C_6_3C_6_4TraceMulA_3_3A_3_4A_3_2A_3_5A_4_3A_4_4A_4_2A_4_5B_3_2B_3_1B_3_5B_3_4B_3_6B_3_3B_2_1B_2_2B_2_4B_2_5B_2_3B_2_6C_1_3C_1_2C_1_1C_1_4C_4_2C_4_3C_4_1C_4_4C_3_2C_3_3C_3_1C_3_4TraceMulA_3_3A_2_3A_2_2A_3_2A_1_3A_4_3A_1_2A_4_2B_3_1B_4_1B_4_4B_3_4B_3_3B_4_3B_2_1B_5_1B_2_4B_5_4B_2_3B_5_3C_1_3C_2_3C_1_4C_2_4C_4_3C_5_3C_5_4C_4_4C_3_3C_6_3C_3_4C_6_4TraceMulA_2_1A_2_4A_2_5A_1_1A_1_4A_1_5B_1_2B_1_5B_1_6B_4_2B_4_5B_4_6B_5_2B_5_5B_5_6C_2_2C_2_1C_5_2C_5_1C_6_2C_6_1TraceMulA_3_1A_3_4A_3_5A_4_1A_4_4A_4_5B_1_1B_1_2B_1_4B_1_5B_1_3B_1_6B_3_1B_4_1B_3_2B_4_2B_4_4B_3_4B_3_5B_4_5B_3_3B_4_3B_4_6B_3_6B_2_1B_5_1B_5_2B_2_2B_5_4B_2_4B_2_5B_5_5B_2_3B_5_3B_2_6B_5_6C_2_3C_2_4C_5_3C_5_4C_6_3C_6_4TraceMulA_3_1A_2_1A_2_3A_3_3A_2_4A_3_4A_3_2A_2_2A_2_5A_3_5A_1_1A_4_1A_1_3A_4_3A_1_4A_4_4A_1_2A_4_2A_1_5A_4_5B_1_1B_1_4B_1_3B_4_1B_4_4B_4_3B_5_1B_5_4B_5_3C_1_2C_1_1C_4_2C_4_1C_3_2C_3_1TraceMulA_2_3A_2_2A_1_3A_1_2B_3_2B_3_5B_3_6B_2_2B_2_5B_2_6C_1_3C_2_3C_2_2C_1_2C_1_1C_2_1C_1_4C_2_4C_4_2C_5_2C_4_3C_5_3C_4_1C_5_1C_4_4C_5_4C_3_2C_6_2C_3_3C_6_3C_3_1C_6_1C_3_4C_6_4TraceMulA_3_1A_2_3A_3_3A_3_4A_3_2A_2_2A_3_5A_4_1A_1_3A_4_3A_4_4A_1_2A_4_2A_4_5B_1_1B_1_4B_1_3B_3_1B_4_1B_3_2B_4_4B_3_4B_3_5B_3_3B_4_3B_3_6B_2_1B_5_1B_2_2B_2_4B_5_4B_2_5B_2_3B_5_3B_2_6C_1_3C_1_2C_2_3C_1_1C_1_4C_2_4C_4_2C_4_3C_5_3C_4_1C_4_4C_5_4C_3_2C_3_3C_6_3C_3_1C_3_4C_6_4Trace(Mul(Matrix(4, 5, [[A_1_1,A_1_2,A_1_3,A_1_4,A_1_5],[A_2_1,A_2_2,A_2_3,A_2_4,A_2_5],[A_3_1,A_3_2,A_3_3,A_3_4,A_3_5],[A_4_1,A_4_2,A_4_3,A_4_4,A_4_5]]),Matrix(5, 6, [[B_1_1,B_1_2,B_1_3,B_1_4,B_1_5,B_1_6],[B_2_1,B_2_2,B_2_3,B_2_4,B_2_5,B_2_6],[B_3_1,B_3_2,B_3_3,B_3_4,B_3_5,B_3_6],[B_4_1,B_4_2,B_4_3,B_4_4,B_4_5,B_4_6],[B_5_1,B_5_2,B_5_3,B_5_4,B_5_5,B_5_6]]),Matrix(6, 4, [[C_1_1,C_1_2,C_1_3,C_1_4],[C_2_1,C_2_2,C_2_3,C_2_4],[C_3_1,C_3_2,C_3_3,C_3_4],[C_4_1,C_4_2,C_4_3,C_4_4],[C_5_1,C_5_2,C_5_3,C_5_4],[C_6_1,C_6_2,C_6_3,C_6_4]]))) = Trace(Mul(Matrix(2, 2, [[A_3_3-A_3_4,A_3_2-A_3_5],[A_4_3-A_4_4,A_4_2-A_4_5]]),Matrix(2, 3, [[-B_3_2+B_3_1,-B_3_5+B_3_4,-B_3_6+B_3_3],[B_2_1-B_2_2,B_2_4-B_2_5,B_2_3-B_2_6]]),Matrix(3, 2, [[C_1_3-C_1_2,-C_1_1+C_1_4],[-C_4_2+C_4_3,-C_4_1+C_4_4],[-C_3_2+C_3_3,-C_3_1+C_3_4]])))+Trace(Mul(Matrix(2, 2, [[A_3_3+A_2_3,A_2_2+A_3_2],[A_1_3+A_4_3,A_1_2+A_4_2]]),Matrix(2, 3, [[B_3_1+B_4_1,B_4_4+B_3_4,B_3_3+B_4_3],[B_2_1+B_5_1,B_2_4+B_5_4,B_2_3+B_5_3]]),Matrix(3, 2, [[C_1_3+C_2_3,C_1_4+C_2_4],[C_4_3+C_5_3,C_5_4+C_4_4],[C_3_3+C_6_3,C_3_4+C_6_4]])))+Trace(Mul(Matrix(2, 3, [[A_2_1,A_2_4,A_2_5],[A_1_1,A_1_4,A_1_5]]),Matrix(3, 3, [[B_1_2,B_1_5,B_1_6],[B_4_2,B_4_5,B_4_6],[B_5_2,B_5_5,B_5_6]]),Matrix(3, 2, [[C_2_2,C_2_1],[C_5_2,C_5_1],[C_6_2,C_6_1]])))+Trace(Mul(Matrix(2, 3, [[A_3_1,A_3_4,A_3_5],[A_4_1,A_4_4,A_4_5]]),Matrix(3, 3, [[-B_1_1+B_1_2,-B_1_4+B_1_5,-B_1_3+B_1_6],[-B_3_1-B_4_1+B_3_2+B_4_2,-B_4_4-B_3_4+B_3_5+B_4_5,-B_3_3-B_4_3+B_4_6+B_3_6],[-B_2_1-B_5_1+B_5_2+B_2_2,-B_5_4-B_2_4+B_2_5+B_5_5,-B_2_3-B_5_3+B_2_6+B_5_6]]),Matrix(3, 2, [[C_2_3,C_2_4],[C_5_3,C_5_4],[C_6_3,C_6_4]])))+Trace(Mul(Matrix(2, 3, [[A_3_1+A_2_1,-A_2_3-A_3_3+A_2_4+A_3_4,-A_3_2-A_2_2+A_2_5+A_3_5],[A_1_1+A_4_1,-A_1_3-A_4_3+A_1_4+A_4_4,-A_1_2-A_4_2+A_1_5+A_4_5]]),Matrix(3, 3, [[B_1_1,B_1_4,B_1_3],[B_4_1,B_4_4,B_4_3],[B_5_1,B_5_4,B_5_3]]),Matrix(3, 2, [[C_1_2,C_1_1],[C_4_2,C_4_1],[C_3_2,C_3_1]])))+Trace(Mul(Matrix(2, 2, [[A_2_3,A_2_2],[A_1_3,A_1_2]]),Matrix(2, 3, [[B_3_2,B_3_5,B_3_6],[B_2_2,B_2_5,B_2_6]]),Matrix(3, 2, [[-C_1_3-C_2_3+C_2_2+C_1_2,C_1_1+C_2_1-C_1_4-C_2_4],[C_4_2+C_5_2-C_4_3-C_5_3,C_4_1+C_5_1-C_4_4-C_5_4],[C_3_2+C_6_2-C_3_3-C_6_3,C_3_1+C_6_1-C_3_4-C_6_4]])))+Trace(Mul(Matrix(2, 3, [[A_3_1,-A_2_3-A_3_3+A_3_4,-A_3_2-A_2_2+A_3_5],[A_4_1,-A_1_3-A_4_3+A_4_4,-A_1_2-A_4_2+A_4_5]]),Matrix(3, 3, [[B_1_1,B_1_4,B_1_3],[B_3_1+B_4_1-B_3_2,B_4_4+B_3_4-B_3_5,B_3_3+B_4_3-B_3_6],[B_2_1+B_5_1-B_2_2,B_2_4+B_5_4-B_2_5,B_2_3+B_5_3-B_2_6]]),Matrix(3, 2, [[C_1_3-C_1_2+C_2_3,-C_1_1+C_1_4+C_2_4],[-C_4_2+C_4_3+C_5_3,-C_4_1+C_4_4+C_5_4],[-C_3_2+C_3_3+C_6_3,-C_3_1+C_3_4+C_6_4]])))

N.B.: for any matrices A, B and C such that the expression Tr(Mul(A,B,C)) is defined, one can construct several trilinear homogeneous polynomials P(A,B,C) such that P(A,B,C)=Tr(Mul(A,B,C)) (P(A,B,C) variables are A,B and C's coefficients). Each trilinear P expression encodes a matrix multiplication algorithm: the coefficient in C_i_j of P(A,B,C) is the (i,j)-th entry of the matrix product Mul(A,B)=Transpose(C).

Algorithm description

These encodings are given in compressed text format using the maple computer algebra system. In each cases, the last line could be understood as a description of the encoding with respect to classical matrix multiplication algorithm. As these outputs are structured, one can construct easily a parser to its favorite format using the maple documentation without this software.


Back to main table