We are left with following problem, upon which TcT provides the certificate YES(O(1),O(n^2)). Strict Trs: { plus_x#1(0(), x8) -> x8 , plus_x#1(S(x12), x14) -> S(plus_x#1(x12, x14)) , map#2(plus_x(x2), Nil()) -> Nil() , map#2(plus_x(x6), Cons(x4, x2)) -> Cons(plus_x#1(x6, x4), map#2(plus_x(x6), x2)) , main(x5, x12) -> map#2(plus_x(x12), x5) } Obligation: innermost runtime complexity Answer: YES(O(1),O(n^2)) We add the following weak dependency pairs: Strict DPs: { plus_x#1^#(0(), x8) -> c_1() , plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , map#2^#(plus_x(x2), Nil()) -> c_3() , map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) , main^#(x5, x12) -> c_5(map#2^#(plus_x(x12), x5)) } and mark the set of starting terms. We are left with following problem, upon which TcT provides the certificate YES(O(1),O(n^2)). Strict DPs: { plus_x#1^#(0(), x8) -> c_1() , plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , map#2^#(plus_x(x2), Nil()) -> c_3() , map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) , main^#(x5, x12) -> c_5(map#2^#(plus_x(x12), x5)) } Strict Trs: { plus_x#1(0(), x8) -> x8 , plus_x#1(S(x12), x14) -> S(plus_x#1(x12, x14)) , map#2(plus_x(x2), Nil()) -> Nil() , map#2(plus_x(x6), Cons(x4, x2)) -> Cons(plus_x#1(x6, x4), map#2(plus_x(x6), x2)) , main(x5, x12) -> map#2(plus_x(x12), x5) } Obligation: innermost runtime complexity Answer: YES(O(1),O(n^2)) No rule is usable, rules are removed from the input problem. We are left with following problem, upon which TcT provides the certificate YES(O(1),O(n^2)). Strict DPs: { plus_x#1^#(0(), x8) -> c_1() , plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , map#2^#(plus_x(x2), Nil()) -> c_3() , map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) , main^#(x5, x12) -> c_5(map#2^#(plus_x(x12), x5)) } Obligation: innermost runtime complexity Answer: YES(O(1),O(n^2)) The weightgap principle applies (using the following constant growth matrix-interpretation) The following argument positions are usable: Uargs(c_2) = {1}, Uargs(c_4) = {1, 2}, Uargs(c_5) = {1} TcT has computed the following constructor-restricted matrix interpretation. [0] = [0] [0] [S](x1) = [1 2] x1 + [0] [0 1] [0] [plus_x](x1) = [1 0] x1 + [0] [0 0] [0] [Nil] = [0] [0] [Cons](x1, x2) = [1 0] x2 + [0] [0 0] [0] [plus_x#1^#](x1, x2) = [0 0] x1 + [0] [1 0] [0] [c_1] = [0] [0] [c_2](x1) = [1 0] x1 + [0] [0 1] [0] [map#2^#](x1, x2) = [0] [0] [c_3] = [0] [0] [c_4](x1, x2) = [1 0] x1 + [1 0] x2 + [0] [0 1] [0 1] [0] [main^#](x1, x2) = [1 2] x1 + [2 2] x2 + [2] [1 2] [2 2] [2] [c_5](x1) = [1 0] x1 + [0] [0 1] [2] The order satisfies the following ordering constraints: [plus_x#1^#(0(), x8)] = [0] [0] >= [0] [0] = [c_1()] [plus_x#1^#(S(x12), x14)] = [0 0] x12 + [0] [1 2] [0] >= [0 0] x12 + [0] [1 0] [0] = [c_2(plus_x#1^#(x12, x14))] [map#2^#(plus_x(x2), Nil())] = [0] [0] >= [0] [0] = [c_3()] [map#2^#(plus_x(x6), Cons(x4, x2))] = [0] [0] ? [0 0] x6 + [0] [1 0] [0] = [c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2))] [main^#(x5, x12)] = [2 2] x12 + [1 2] x5 + [2] [2 2] [1 2] [2] > [0] [2] = [c_5(map#2^#(plus_x(x12), x5))] Further, it can be verified that all rules not oriented are covered by the weightgap condition. We are left with following problem, upon which TcT provides the certificate YES(O(1),O(n^2)). Strict DPs: { plus_x#1^#(0(), x8) -> c_1() , plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , map#2^#(plus_x(x2), Nil()) -> c_3() , map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) } Weak DPs: { main^#(x5, x12) -> c_5(map#2^#(plus_x(x12), x5)) } Obligation: innermost runtime complexity Answer: YES(O(1),O(n^2)) We estimate the number of application of {1} by applications of Pre({1}) = {2,4}. Here rules are labeled as follows: DPs: { 1: plus_x#1^#(0(), x8) -> c_1() , 2: plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , 3: map#2^#(plus_x(x2), Nil()) -> c_3() , 4: map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) , 5: main^#(x5, x12) -> c_5(map#2^#(plus_x(x12), x5)) } We are left with following problem, upon which TcT provides the certificate YES(O(1),O(n^2)). Strict DPs: { plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , map#2^#(plus_x(x2), Nil()) -> c_3() , map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) } Weak DPs: { plus_x#1^#(0(), x8) -> c_1() , main^#(x5, x12) -> c_5(map#2^#(plus_x(x12), x5)) } Obligation: innermost runtime complexity Answer: YES(O(1),O(n^2)) The following weak DPs constitute a sub-graph of the DG that is closed under successors. The DPs are removed. { plus_x#1^#(0(), x8) -> c_1() } We are left with following problem, upon which TcT provides the certificate YES(O(1),O(n^2)). Strict DPs: { plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , map#2^#(plus_x(x2), Nil()) -> c_3() , map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) } Weak DPs: { main^#(x5, x12) -> c_5(map#2^#(plus_x(x12), x5)) } Obligation: innermost runtime complexity Answer: YES(O(1),O(n^2)) Consider the dependency graph 1: plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) -->_1 plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) :1 2: map#2^#(plus_x(x2), Nil()) -> c_3() 3: map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) -->_2 map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) :3 -->_2 map#2^#(plus_x(x2), Nil()) -> c_3() :2 -->_1 plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) :1 4: main^#(x5, x12) -> c_5(map#2^#(plus_x(x12), x5)) -->_1 map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) :3 -->_1 map#2^#(plus_x(x2), Nil()) -> c_3() :2 Following roots of the dependency graph are removed, as the considered set of starting terms is closed under reduction with respect to these rules (modulo compound contexts). { main^#(x5, x12) -> c_5(map#2^#(plus_x(x12), x5)) } We are left with following problem, upon which TcT provides the certificate YES(O(1),O(n^2)). Strict DPs: { plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , map#2^#(plus_x(x2), Nil()) -> c_3() , map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) } Obligation: innermost runtime complexity Answer: YES(O(1),O(n^2)) We estimate the number of application of {2} by applications of Pre({2}) = {3}. Here rules are labeled as follows: DPs: { 1: plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , 2: map#2^#(plus_x(x2), Nil()) -> c_3() , 3: map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) } We are left with following problem, upon which TcT provides the certificate YES(O(1),O(n^2)). Strict DPs: { plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) } Weak DPs: { map#2^#(plus_x(x2), Nil()) -> c_3() } Obligation: innermost runtime complexity Answer: YES(O(1),O(n^2)) The following weak DPs constitute a sub-graph of the DG that is closed under successors. The DPs are removed. { map#2^#(plus_x(x2), Nil()) -> c_3() } We are left with following problem, upon which TcT provides the certificate YES(O(1),O(n^2)). Strict DPs: { plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) } Obligation: innermost runtime complexity Answer: YES(O(1),O(n^2)) We use the processor 'Small Polynomial Path Order (PS,2-bounded)' to orient following rules strictly. DPs: { 1: plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , 2: map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) } Sub-proof: ---------- The input was oriented with the instance of 'Small Polynomial Path Order (PS,2-bounded)' as induced by the safe mapping safe(S) = {1}, safe(plus_x) = {1}, safe(Cons) = {1, 2}, safe(plus_x#1^#) = {2}, safe(c_2) = {}, safe(map#2^#) = {}, safe(c_4) = {} and precedence map#2^# > plus_x#1^# . Following symbols are considered recursive: {plus_x#1^#, map#2^#} The recursion depth is 2. Further, following argument filtering is employed: pi(S) = [1], pi(plus_x) = 1, pi(Cons) = [2], pi(plus_x#1^#) = [1], pi(c_2) = [1], pi(map#2^#) = [1, 2], pi(c_4) = [1, 2] Usable defined function symbols are a subset of: {plus_x#1^#, map#2^#} For your convenience, here are the satisfied ordering constraints: pi(plus_x#1^#(S(x12), x14)) = plus_x#1^#(S(; x12);) > c_2(plus_x#1^#(x12;);) = pi(c_2(plus_x#1^#(x12, x14))) pi(map#2^#(plus_x(x6), Cons(x4, x2))) = map#2^#(x6, Cons(; x2);) > c_4(plus_x#1^#(x6;), map#2^#(x6, x2;);) = pi(c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2))) The strictly oriented rules are moved into the weak component. We are left with following problem, upon which TcT provides the certificate YES(O(1),O(1)). Weak DPs: { plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) } Obligation: innermost runtime complexity Answer: YES(O(1),O(1)) The following weak DPs constitute a sub-graph of the DG that is closed under successors. The DPs are removed. { plus_x#1^#(S(x12), x14) -> c_2(plus_x#1^#(x12, x14)) , map#2^#(plus_x(x6), Cons(x4, x2)) -> c_4(plus_x#1^#(x6, x4), map#2^#(plus_x(x6), x2)) } We are left with following problem, upon which TcT provides the certificate YES(O(1),O(1)). Rules: Empty Obligation: innermost runtime complexity Answer: YES(O(1),O(1)) Empty rules are trivially bounded Hurray, we answered YES(O(1),O(n^2))