1:
2:
3:
4:
5:
6:
7: \section{Analyzing Goal Asymmetry in Synthetic Domains}
8: \label{sd}
9:
10:
11:
12:
13:
14: In this section, we perform a number of case studies. We analyze
15: synthetic domains constructed explicitly to provoke interesting
16: behavior regarding $\ASratio$. The aim of the analysis is to obtain a
17: better understanding of how this sort of problem structure affects the
18: behavior of SAT solvers; in fact, the definition of $\ASratio$ was
19: motivated in the first place by observations we made in synthetic
20: examples.
21:
22:
23:
24: The analytical results we obtain in our case studies are, of course,
25: specific to the studied domains. We do, however, identify a set of
26: prototypical patterns of structure that also appear in the planning
27: competition examples; we will point this out in the text.
28:
29:
30:
31: We analyze three classes of synthetic domains/CNF formulas, called
32: MAP, SBW, and SPH. MAP is a simple transportation kind of domain, SBW
33: is a block stacking domain. SPH is a structured version of the pigeon
34: hole problem. Each of the domains/CNF classes is parameterized by size
35: $n$ and structure $k$. In the planning domains, we use the simplified
36: Graphplan-based encoding described in Section~\ref{prelim}, and
37: consider CNFs that are one step short of a solution. We denote the
38: CNFs with $MAP_n^k$, $SBW_n^k$, and $SPH_n^k$, respectively.
39:
40:
41:
42: We choose the MAP and SBW domains because they are related to
43: Logistics and Blocksworld, two of the most classical Planning
44: benchmarks. We chose SPH for its close relation to the formulas
45: considered in proof complexity. The reader will notice that the
46: synthetic domains are {\em very} simple. The reasons for this are
47: threefold. First, we wanted to capture the intended intuitive problem
48: structure in as clean a form as possible, without ``noise''. Second,
49: even though the Planning tasks are quite simple, the resulting CNF
50: formulas are complicated -- e.g., much more complicated than the
51: pigeon hole formulas often considered in proof complexity. Third, we
52: identify provably minimal backdoors. To do so, one has to take
53: account of every tiny detail of the effects of unit propagation. The
54: respective proofs are already quite involved for our simple domains --
55: for MAP, e.g., they occupy 9 pages, featuring myriads of interleaved
56: case distinctions. To analyze more complex domains, one probably has
57: to sacrifice precision. \ifTR{For the sake of readability, the proofs
58: are moved to Appendix~\ref{proofs}, and only briefly sketched here.}
59:
60:
61: \ifLMCS{For the sake of readability, herein we discuss only MAP in
62: detail, and we replace the proofs with proof sketches. The details for
63: SBW and SPH, and the proofs, are in the TR
64: \cite{hoffmann:etal:tr-lmcs06}.}
65:
66:
67:
68:
69:
70: \input{MAP}
71:
72: \ifTR{
73: \input{SBW}
74: \input{SPH}
75: }
76:
77:
78: \ifLMCS{
79:
80: \subsection{SBW}
81: \label{sd:sbw}
82:
83:
84: This is a block-stacking domain, with stacking restrictions on what
85: blocks can be stacked onto what other blocks. The blocks are
86: initially all located side-by-side on a table $t_1$. The goal is to
87: bring all blocks onto another table $t_2$, that has only space for a
88: single block; so the $n$ blocks must be arranged in a single stack on
89: top of $t_2$. The parameter $k$, $0 \leq k \leq n$, defines the amount
90: of restrictions. There are $k$ ``bad'' blocks $b_1, \dots, b_k$ and
91: $n-k$ ``good'' blocks $g_1, \dots, g_{n-k}$. Each $b_i$, $i>1$, can
92: only be stacked onto $b_{i-1}$; $b_1$ can be stacked onto $t_2$ and
93: any $g_i$. The $g_i$ can be stacked any $g_j$, and onto $t_2$.
94:
95: Independently of $k$, the optimal plan length is $n$: move actions
96: stack one block onto another block or a table. $\ASratio$ is $1/n$ if
97: $k=0$, and $k/n$ otherwise. In the symmetrical case, $k=0$, we
98: identify backdoors of size $\Theta(n^3)$ -- linear in the total number
99: of variables. In the asymmetrical case, $k=n-2$, there are $O(log n)$
100: DPLL refutations and backdoors. It is an open question whether there
101: is an exponential lower bound in the symmetrical case.
102:
103:
104:
105:
106:
107:
108:
109:
110: \subsection{SPH}
111: \label{sd:sph}
112:
113:
114:
115: For this domain, we modified the pigeon hole problem. In our $SPH_n^k$
116: formulas, as usual the task is to assign $n+1$ pigeons to $n$
117: holes. The new feature is that there is one ``bad'' pigeon that
118: requires $k$ holes, and $k-1$ ``good'' pigeons that can share a hole
119: with the bad pigeon. The remaining $n-k+1$ pigeons are normal, i.e.,
120: need exactly one hole each. The range of $k$ is between $1$ and $n-1$.
121: Independently of $k$, $n+1$ holes are needed overall.
122:
123: We identify minimal backdoors for all combinations of $k$ and $n$;
124: their size is $(n-k)*(n-1)$. For $k = n-1$ we identify an $O(n)$ DPLL
125: refutation. With results by Buss and Pitassi
126: \cite{buss:pitassi:csl-97}, this implies an exponential DPLL
127: complexity gap to $k = 1$.
128:
129: }
130:
131:
132:
133:
134:
135: