For a polynomial, use the sum of absolute coefficients; for a matrix, use the largest absolute row sum. Both bounds are submultiplicative, by expanding products and applying the triangle inequality. The construction gives ||W||<=Q, ||M_0||<=R*(4Q)^(R-1), and ||phi||<=2^Q. Hence a cleared row has total absolute coefficient sum at most delta*[1+(1+2^Q)*R*(4Q)^(R-1)]. With 1<=R<Q, its base-two logarithm is O(Q*log Q), so each coefficient needs that many bits. Copy filling cannot increase the total absolute coefficient sum. Enumerating and manipulating arrays polynomial in D,Q,r therefore has polynomial bit overhead when these parameters are polynomially bounded.