Define d(x,y) = d(H(x),H(y)) where H(x) denotes the sequence (H_q(x,eps)) for some eps fixed for now. Say d(H(x),H(y)) = sum_q d(H_q(x),H_q(y)). Recall H_q(x) = z + B_q(x) for all z in Z_q(x). One way to define d(H_q(x),H_q(y)) is as |beta_q(x)-beta_q(y)| i.e. |rank(H_q(x))-rank(H_q(y))|. The problem is that this ignores which cycles were involved in the cycles. Idea: Just like we quotient out the cycles which are also boundaries, let's also quotient out the cycles which are common. For example d-----b versus b-----f \ / \ / \***/ \ / \ / \*/ c-----a c-----a x y gives C_0(x) = Z_0(x) = Ker (del_0 : C_0 -> C_{-1}) = C_0(x) B_0(x) = Im (del_1 : C_1 -> C_0) = {0} H_0(x) = Z_0(x)/B_0(x) = C_0(x)/B_0(x) = /{0} = C_0(y) = Z_0(y) = Ker (del_0 : C_0 -> C_{-1}) = C_0(x) B_0(y) = Im (del_1 : C_1 -> C_0) = {0} H_0(y) = Z_0(x)/B_0(x) = C_0(x)/B_0(x) = /{0} = C_1(x) = Z_1(x) = Ker (del_1 : C_1 -> C_0) = B_1(x) = Im (del_2 : C_2 -> C_1) = Im (del_2 : {0} -> C_1) = {0} H_1(x) = Z_1(x)/B_1(x) = /{0} = C_1(y) = Z_1(y) = Ker (del_1 : C_1 -> C_0) = B_1(y) = Im (del_2 : C_2 -> C_1) = Im (del_2 : -> C_1) = H_1(y) = Z_1(y)/B_1(y) = / = ,ab+bc+ca+> = <0+,ab+bc+ca+> = > Now we want to throw out the ab+bc+ca cycles because they are in common. So we define Z'_1(x) = Z_1(x)/(Z_1(x) xsect Z_1(y)) = /( xsect ) = / = > B'_1(x) = B_1(x)/(Z_1(x) xsect Z_1(y)) = {0}/ = {0+} H'_1(x) = Z'_1(x)/B'_1(x) = >/{0+} = > = > Z'_1(y) = Z_1(y)/(Z_1(x) xsect Z_1(y)) = / = > B'_1(y) = B_1(y)/(Z_1(x) xsect Z_1(y)) = / = > H'_1(y) = Z'_1(y)/B'_1(y) = >/> = <(af+fb+ba+) + > > = <0 + > > ~= {0}