安德烈(Andrej)已经解释说 P=P小号P一个CËP=PSPACE 可以写成 Σ02Σ02-句子。我要说的是,这种分类是最佳的,即如果该语句等于一个Π02Π02-句子,那么这个事实就不会相对化。更确切地说,这套神谕一个A 这样 P一个=P小号P一个CË一个PA=PSPACEA 由一个定义 Σ02Σ02带有自由二阶变量的公式 一个A, but it is not definable by any Π02Π02-formula. The argument is outlined (for P=NPP=NP, but it works just the same for PSPACEPSPACE) in the comments at /mathpro/57348. (In fact, one can show by an elaboration of the idea that the set is Σ02Σ02-complete in the appropriate sense.)
EDIT: The topological proof given in the linked comment is short, but it may appear tricky. Here is a direct forcing argument.
PA≠PSPACEAPA≠PSPACEA can be written as a Π02Π02-formula of the form ϕ(A)=∀x∃yθ(A,x,y)ϕ(A)=∀x∃yθ(A,x,y), where θθ is Δ00Δ00. Assume for contradiction that PA=PSPACEAPA=PSPACEA is also equivalent to a Π02Π02-formula ψ(A)=∀x∃zη(A,x,z)ψ(A)=∀x∃zη(A,x,z). Fix oracles BB, CC such that PB≠PSPACEBPB≠PSPACEB and PC=PSPACECPC=PSPACEC.
Since ϕ(B)ϕ(B), there exists y0y0 such that θ(B,0,y0)θ(B,0,y0). However, θθ is a bounded formula, hence the evaluation of the truth value of θ(B,0,y0)θ(B,0,y0) only uses a finite part of the oracle. Thus, there exists a finite part b0b0 of BB such that θ(A,0,y0)θ(A,0,y0) for every oracle AA extending b0b0.
Let C[b0]C[b0] denote the oracle which extends b0b0, and agrees with CC where b0b0 is undefined. Since PAPA and PSPACEAPSPACEA are unaffected by a finite change in the oracle, we have ψ(C[b0])ψ(C[b0]). By the same argument as above, there exists z0z0 and a finite part c0c0 of C[b0]C[b0] such that η(A,0,z0)η(A,0,z0) for every AA extending c0c0. We may assume that c0c0 extends b0b0.
Continuing in the same fashion, we construct infinite sequences of numbers y0,y1,y2,…, z0,z1,z2,…, and finite partial oracles b0⊆c0⊆b1⊆c1⊆b2⊆⋯ such that
θ(A,n,yn) for every oracle A extending bn,
η(A,n,zn) for every oracle A extending cn.
Now, let A be an oracle which extends all bn and cn. Then 1 and 2 imply that ϕ(A) and ψ(A) simultaneously hold, which contradicts the assumption that they are complements of each other.