将PSPACE与多项式层次结构区分开的最小复杂度预告片是什么?


17

背景

已知的是,存在一个预言使得。P S P A C E AP H AAPSPACEAPHA

甚至众所周知,相对于随机预言片而言,这种分离成立。非正式地讲,这可能意味着有许多预言将PSPACEPH分开。

这些将PSPACE分开的预言有多复杂PH。特别地,有一个oracle ADTIME(22n),使得 PSPACEAPHA

我们是否有任何预言A使得PSPACEAPHAA具有已知的复杂度上限?

注意:这种预言的存在可能会对结构复杂性理论产生影响。有关更多详细信息,请参见下面的更新。

更新有关下限技术的详细信息

权利要求:如果,那么对于所有的预言P / p ø ÿP 小号P Ç é = P ħ PSPACE=PHAP/polyPSPACEA=PHA

证明示意图:假设PSPACE=PH

让一个oracle 给出。我们可以建立一个多项式时间 Σ 2预言图灵机中号,对于给定长度Ñ,猜测的尺寸的电路p Ñ 使用存在量化并验证该电路判定通过比较电路的评估和查询结果对于每个长度n个字符串,使用通用量化。AP/polyΣ2Mnp(n)An

此外,考虑一个决策问题,我将其称为量化布尔电路(QBC),您将获得一个量化布尔电路,并想知道它是否有效(类似于QBF)。此问题是PSPACE完整的,因为QBF是PSPACE完整的。

由假设,它遵循QBC 。比方说,Q Ç ∈ Σ ķ一些ķ足够大。设Ñ表示一个多项式时间Σ ķ图灵机可以解决QBC。PHQBCΣkkNΣk

我们可以交融的计算ñ(类似于在卡普,立顿定理的证明进行),以获得一个多项式时间 Σ ķ甲骨文图灵机,解决了Q Ç 一个MNΣkQBCA

非正式地,这台新机器将oracle QBC(即带有oracle gates的QBC)作为输入。然后,它计算一个电路,该电路在长度为n的输入上计算 (同时剔除前两个量词)。接下来,它将A的电路替换为Oracle QBC中的Oracle门。最后,它继续在此修改实例上应用多项式时间k算法的其余部分来求解Q B C。AnAΣkQBC

现在,我们可以显示条件下界。

推论:如果存在一个oracle ,使得P 小号P Ç é P ħ ,然后Ñ Ë X P P / p ö ÿANEXPPSPACEAPHANEXPP/poly

证明示意图:假设存在,使得P 小号P Ç é P ħ 。如果ñ Ë X P P / p Ø Ÿ,那么我们将得到一个矛盾。ANEXPPSPACEAPHANEXPP/poly

特别是,如果,然后由如权利要求上面我们有P 小号P Ç é P ħ。然而,已知的是,Ñ Ë X P P / p ö ý意味着P 小号P Ç é = P ħNEXPP/polyPSPACEPHNEXPP/polyPSPACE=PH

(有关P / poly已知结果的一些详细信息,请参见此处


3
值得一提的是,有人推测PSPACE PH。即一个简单的预言家会做,但是我们无法证明这一点。
托马斯(Thomas)支持莫妮卡(Monica)

1
究竟如何定义相对化的PSPACE?文献中出现了不止一种可能性。特别是,是否假定Oracle查询是多项式有界的?
EmilJeřábek

1
您是否在PH中加入了“ Q公式的构造”,即决定原始公式的所有2 ^ n qbfs的大型单调布尔型公式?有关Q公式的更多信息,请参见2002年可满足性会议QBFS国际研讨会上的QSpace简介。
丹尼尔·皮侯谢克

1
我相信我可以显示,作为下限,这样的在被SEH会“在结构复杂性理论的后果。” 我是否应该尽快发布(可能意味着明天或可能要在30分钟后发布),还是将其保留更长的时间,以便您更有可能获得足够的答案?A

1
考虑到随机预言片具有很高的Kolmogorov复杂度,我希望此类预言片上任何可计算的上限都会产生明显的后果。强大的上限(如单指数)应具有强大的后果。(当然,这种说法纯粹是试探性的,我目前不知道如何使其严格。)
AndrásSalamon

Answers:


9

我相信,如果您追溯到例如Ker-I Ko的调查的 4.1节中给出的论点,您将获得的上限。实际上,我们可以在这里用任何函数n f n 替换n 2,其中f n n 。这不是要求的,但是已经很接近了。DTIME(22O(n2))n2nf(n)f(n)n

特别地,使用oracle分隔和电路下限之间的转换并遵循Ko的表示法,我们具有以下内容:

  • 我们将对长度为字符串进行对角线化,其中p nx = x n + n是“第n个多项式”(在多项时间算法的一些枚举中),m n 将在下面指定。t(n)=pn(m(n))pn(x)=xn+nnm(n)

  • Translating into circuit lower bounds, this means we're considering bounded-depth circuits on 2t(n) inputs.

  • The requirement (see p. 15 of Ko) we need m(n) to satisfy 1102m/(d1)>dpn(m(n)) for all n. Here d is the depth of the circuits we want to diagonalize against, or equivalently the level Σdp of PH we want to diagonalize against. To diagonalize against all of PH, simply choose d to be a function of n that is ω(1); we may choose such a d that grows arbitrarily slowly, though (perhaps subject to some computability assumption on d(n), but that should be no obstacle). If we make the guess that d(n) is constant (even though it's not, but it will grow arbitrarily slowly), then we see that m(n) around 2n should work.

  • This means that t(n)2n2, so we are looking for a lower bound against circuits with 22n2 inputs.

  • Trevisan and Xue (CCC '13) showed that one can find an assignment on which a given bounded-depth circuit on N inputs doesn't compute PARITY with a seed of polylog(N) length.

  • For us N=22n2, so polylog(N)=2O(n2). We can brute force over such seeds in 22O(n2) time and use the first one that works.

To replace the n2 with nf(n), just let pn(x)=xf(n)+f(n) instead.

Interestingly, if I'm understanding correctly, I believe this implies that if one could improve the Trevisan-Xue...

  • ...to a pseudodeterministic/Bellagio algorithm (see Andrew Morgan's comment below), one would get that BPEXPP/poly; or

  • polylog(N) bits but then ran in poly(N) time, and such that on any accepting path it makes the same output (cf. NPSV), it would imply NEXPP/poly; or

  • ... to a deterministic algorithm, one would get EXPP/poly.

On the one hand, this suggests why derandomizing the switching lemma further should be hard - an argument which I'm not sure was known before! On the other hand, this strikes me as a kind of interesting take on hardness versus randomness (or is this actually a new thing, oracles versus randomness?).


3
One challenge that's glossed over here is that the oracle that's constructed has to be a single, fixed oracle, so that deciding it is in BPEXP or whatever. If you just pick a random seed of a good generator, then, while you do get some oracle that works, you don't necessarily get a decision procedure for that oracle, since different seeds give (in general) different oracles. You'd have to do something more, like finding a canonical seed, in order to make the construction actually "constructive".
Andrew Morgan

3
Even though the argument doesn’t give BPEXP, can you get the complexity down to a finite level of EXPH?
Emil Jeřábek supports Monica

2
@EmilJeřábek: Without checking the details, I think Σ3EXP should work. Guess a seed using , verify it works using , and then verify that it is the lexicographically least seed using ¬=¬, for a total of .
Joshua Grochow

2
@EmilJerabek: Of course, if we could at least get this down to MAEXP that would be even better (not improvable without proving new circuit lower bounds), but I don't yet see how to do that...
Joshua Grochow

2
@JoshuaGrochow Yeah, your original post seems fine. I was objecting to your reply to Emil that hypothesized the oracle can be made in EXPH, where the running time is singly-exponential. In retrospect I should have been more clear about that.
Andrew Morgan
By using our site, you acknowledge that you have read and understand our Cookie Policy and Privacy Policy.
Licensed under cc by-sa 3.0 with attribution required.