Respuesta simple: Para cada problema E X P T I M E - h a r d hay una constante c tal que si pudiéramos resolver el problema en N T I M E ( 2 o ( n 1EXPTIMEhardcc )), entoncesP≠NP.NTIME(2o(n1c))P≠NP
Nota: La constante c proviene de las ampliaciones de tamaño de instancia que resultan de las reducciones.c
Justificación: Sea X un problema E X P T I M E - h a r d . Eso significa que todos los problemas de E X P T I M E es polinomio irreducible tiempo para X . De hecho, podemos mostrar más.XEXPTIMEhardEXPTIMEX
El problema de aceptación para 2 n tiempo limitado máquinas de Turing deterministas está en D T I M E ( n ⋅ 2 n ) ⊆ E X P T I M E y por lo tanto es reducible tiempo polinómico a X .2nDTIME(n⋅2n)⊆EXPTIMEX
Por lo tanto, debe haber alguna constante fija c de modo que cada problema en D T I M E ( 2 n ) sea polinomial reducible en tiempo a X donde el tamaño de instancia es O ( n c ) . Eso es, los casos de tamaño n se reducen a los casos de tamaño O ( n c ) para X .cDTIME(2n)XO(nc)O(nc)X
Ahora, si tuviéramos X ∈ N T I M E ( 2 o ( n 1c))X∈NTIME(2o(n1c)), then DTIME(2n)⊆NTIME(2o(n))DTIME(2n)⊆NTIME(2o(n)). However, this implies P≠NPP≠NP (see below for details).
Additional Details: One can show that P=NPP=NP ⇔⇔ ∃c′∃c′ ∀k∀k NTIME(nk)⊆DTIME(nc′k)NTIME(nk)⊆DTIME(nc′k).
In other words, if you can solve an NPNP-completecomplete problem in polynomial time, then there is a uniform way of speeding up any problem in NPNP.
Now, let's suppose that P=NPP=NP. By the preceding (with kk=1) we get a constant c′c′ such that
NTIME(n)⊆DTIME(nc′).
NTIME(n)⊆DTIME(nc′).
Next, we can use padding to scale up this inclusion and get
NTIME(2n)⊆DTIME(2c′n).
NTIME(2n)⊆DTIME(2c′n).
Then, by the deterministic time hierarchy theorem, we have
NTIME(2n)⊆DTIME(2c′n)⊊DTIME(2(c′+ϵ)n)
NTIME(2n)⊆DTIME(2c′n)⊊DTIME(2(c′+ϵ)n)
for any
ϵ>0ϵ>0.
Therefore, we couldn't have
DTIME(2(c′+ϵ)n)⊆NTIME(2n).DTIME(2(c′+ϵ)n)⊆NTIME(2n).
Further, we couldn't have DTIME(2n)⊆NTIME(2o(n))DTIME(2n)⊆NTIME(2o(n)) because by padding we would get DTIME(2(c′+ϵ)n)⊆NTIME(2o(n))DTIME(2(c′+ϵ)n)⊆NTIME(2o(n)).
Further Question: Does anyone have any simple examples of EXPTIMEEXPTIME-completecomplete problems where we can easily determine the instance size blow-up constant cc?