1: \documentclass[A4,11pt]{article}
2: %%\documentclass[11pt]{llncs}
3: %% \def\DEF{\stackrel{\rm def}{=}}
4:
5: %% \newtheorem{theo}{Theorem}
6: %% \newtheorem{prop}{Proposition}
7: %% \newtheorem{fact}[theo]{Fact}
8: %% \newcommand{\eps}{\varepsilon}
9: %% \newcommand{\para}{\medskip\noindent}
10: %% \usepackage{latexsym,graphicx}
11:
12: %%\begin{document}
13: %% \title{\bf Strip Packing vs. Bin Packing } %\thanks{Supported in part by }}
14: %% \titlerunning{square packing}
15:
16: %% \author{Xin Han \inst{1} Kazuo Iwama \inst{1} Deshi Ye \inst{2}
17: %% Guochuan Zhang \inst{3}\thanks{Research supported by NSFC
18: %% (10231060)} }
19: %% %
20: %% \authorrunning{xin han et al.} % abbreviated author list (for running head)
21: %% \institute{ School of Informatics, Kyoto University, Kyoto
22: %% 606-8501, Japan, \\
23: %% \email{\{hanxin, iwama\}@kuis.kyoto-u.ac.jp},
24: %% \and Department of Computer Science, The University of Hong Kong, Hong Kong,\\
25: %% \email{ yedeshi@cs.hku.hk}
26: %% \and Department of Mathematics, Zhejiang University, China,\\
27: %% \email{zgc@zju.edu.cn}}
28: %% \maketitle % typeset the title of the contribution
29:
30: \setlength{\headheight}{0.00in}
31: \setlength{\footskip}{0.50in}
32:
33: \usepackage{amsfonts}
34: \usepackage{amsmath}
35:
36: %% %%\topmargin -11.0mm \textheight 233.5mm \textwidth 173.0mm
37: \topmargin -9.0mm \textheight 235mm \textwidth 160.0mm
38: \setlength{\oddsidemargin}{-6.0mm}
39: \setlength{\evensidemargin}{-6.0mm}
40:
41: \def\DEF{\stackrel{\rm def}{=}}
42:
43: \newtheorem{lemma}{Lemma}
44: \newtheorem{theorem}{Theorem}%[section]
45: \newenvironment{proof}%
46: {\begin{trivlist}\item[\hspace*{\labelsep}{\it Proof.\/}]}%
47: {\hfill$\Box$\end{trivlist}}
48: \newtheorem{theo}{Theorem}
49: \newtheorem{defi}[theo]{Definition}
50: \newtheorem{prop}{Proposition}
51: \newtheorem{corollary}{Corollary}
52: \newtheorem{obse}[theo]{Observation}
53: \newtheorem{claim}[theo]{Claim}
54: \newtheorem{remark}{Remark}
55: \newtheorem{fact}[theo]{Fact}
56:
57: %\newtheorem{conj}[theo]{Conjecture}
58: %\newcommand{\qed}{\hfill\rule{1.8mm}{1.8mm}}
59: %%\newcommand{\proof}{{\bf Proof.}\ \ }
60: %\newcommand{\rz}{\mbox{{\rm I$\!$R}}}
61: %\newcommand{\Z}{{\bbbz}}
62: %\newcommand{\rz}{\mathbb{Z}}
63: %\newcommand{\nz}{\mbox{{\rm I$\!$N}}}
64: %\newcommand{\eps}{\varepsilon}
65: %% \newcommand{\al}{\alpha}
66: %% \newcommand{\seq}[1]{\langle #1\rangle}
67: %% \newcommand{\flo}[1]{\lfloor #1 \rfloor}
68: %% \newcommand{\cei}[1]{\lceil #1 \rceil}
69: %% \newcommand{\np}{\mbox{${\cal N\cal P}$}}
70: %% \newcommand{\p}{\mbox{${\cal P}$}}
71: %% \newcommand{\para}{\medskip\noindent}
72: %% \newcommand{\bbb}{{\cal B}}
73: %% \newcommand{\lll}{{\cal L}}
74: %% \newcommand{\iii}{{\cal I}}
75: %% \newcommand{\opt}{\mbox{\rm OPT}}
76: %% \newcommand{\head}[1]
77: %% {\markright{\hbox to 0pt{\vtop to 0pt{\hbox{}\vskip 3mm \hrule
78: %% width \textwidth \vss} \hss}{\sc #1}}}
79: \usepackage{latexsym,graphicx}
80: \begin{document}
81: \title{\bf Improved Online Hypercube Packing } %\thanks{Supported in part by }}
82:
83: \author{Xin Han$^1$ \hspace{3mm} Deshi Ye$^2$ \hspace{3mm} Yong Zhou$^3$
84: %%$^3$\thanks{Supported in part by the DFG Project AL
85: %%464/4-1, Eu-Project APPOL II and NSFC (10231060).}
86: \\ {\small $^1$ School of Informatics, Kyoto University, Kyoto
87: 606-8501, Japan} \\ {\small hanxin@kuis.kyoto-u.ac.jp}
88: \\ {\small $^2$ College of Computer Science,
89: Zhejiang University, Hangzhou, 310027, China}
90: \\ {\small yedeshi@zju.edu.cn }
91: \\ {\small $^3$ Graduate School of Science, Hokkaido University, Sapporo, Japan}
92: \\ {\small zhou@castor.sci.hokudai.ac.jp}
93: %%\\ {\small Georges-K\"ohler-Allee 79, 79110 Freiburg, Germany}
94: %% \\ {\small $^3$ Department of Mathematics, Zhejiang University, China}
95: %% \\ {\small zgc@zju.edu.cn}
96: }
97: \date{}
98: \maketitle
99:
100: \baselineskip 13.4pt
101:
102: \begin{abstract}
103: In this paper, we study online multidimensional bin packing problem
104: when all items are hypercubes.
105: %%We propose a new algorithm for hypercube packing,
106: Based on the techniques in one dimensional bin packing
107: algorithm Super Harmonic by Seiden,
108: we give a framework for online hypercube packing problem
109: and obtain new upper bounds of asymptotic competitive ratios.
110: For square packing, we get an upper bound of 2.1439,
111: which is better than 2.24437.
112: For cube packing, we also give a new upper bound 2.6852
113: which is better than 2.9421 by Epstein and van Stee.
114:
115: \end{abstract}
116:
117: %%\baselineskip 12.9pt
118: \section{Introduction}
119: The classical one-dimensional
120: Bin Packing is one of the oldest and most well-studied problems
121: in computer science \cite{CGJ97}, \cite{cw97}.
122: In the early 1970's it was one of the first combinatorial optimization
123: problems for which the idea of worst-case performance guarantees was
124: investigated. It was also in this domain that
125: the idea of proving lower bounds on the performance of online algorithm
126: was first developed.
127: In this paper, we consider a generalization of the
128: classical bin packing problem: hypercube packing problem.
129:
130: \paragraph{\bf Problem Definition.} Let $d \ge 1$ be an integer.
131: %%In the $d$-dimensional bin packing problem
132: We receive a sequence $\delta$ of items $p_1,p_2,...,p_n$.
133: Each item $p$ is a $d$-dimensional hypercube and has a fixed size,
134: which is $s(p)\times \cdots \times s(p)$,
135: i.e., $s(p)$ is the size of $p$ in any dimension.
136: We have an infinite number of bins,
137: each of which is a $d$-dimensional unit hypercube.
138: Each item must be assigned to a position $(x_1(p),...,x_d(p))$ of some bin,
139: where $0 \le x_i(p)$ and
140: $x_i(p) + s(p) \le 1$ for $1 \le i \le d$.
141: Further, the positions must be assigned in such a way
142: that no two items in the same bin overlap.
143: Note that for $d=1$ the problem reduces to the
144: classic bin packing problem.
145: In this paper, we study the {\em online} version of this problem,
146: i.e., each item must be assigned in turn,
147: without knowledge of the next items.
148:
149:
150: \paragraph {\bf Asymptotic competitive ratio.}
151: To evaluate an online algorithms for bin packing,
152: we use the standard measure {\em Asymptotic competitive ratio}
153: which is defined as follows.
154:
155: Given an input list {\em L} and an online algorithm
156: $A$, we denote by $OPT(L)$ and $A(L)$, respectively, the cost
157: (number of bins used) by an optimal (offline) algorithm and the cost
158: by online algorithm $A$ for packing list $L$.
159: The {\em asymptotic competitive ratio} $R_A^{\infty}$
160: of algorithm $A$ is defined by
161: \[
162: R_A^{\infty} =\lim_{k \to \infty} \sup_{L}\{ A(L)/OPT(L)| OPT(L) = k\}.
163: \]
164:
165: \paragraph{\bf Previous results.}
166: On the classic online bin packing,
167: Johnson, Demers, Ullman, Garey and Graham \cite{JDUGG74}
168: showed that the First Fit algorithm has the competitive ratio 1.7.
169: Yao \cite{Yao80} gave an upper bound of 5/3.
170: Lee and Lee \cite{LL85} showed the Harmonic algorithm has
171: the competitive ratio 1.69103 and improved it to 1.63597.
172: Ramanan, Brown, Lee and Lee \cite{RBLL89} improved the upper bound to
173: 1.61217. Currently,
174: the best known upper bound is 1.58889 by Seiden \cite{s02}.
175: On the lower bounds,
176: Yao \cite{Yao80} showed no online algorithm has performance
177: ratio less that 1.5.
178: Brown \cite{B79} and Liang \cite{L80} independently improved
179: this lower bound to 1.53635.
180: The lower bound currently stands at 1.54014, due to van Vliet \cite{Vliet92}.
181:
182:
183: On online hypercube packing,
184: Coppersmith and Raghavan \cite{CR89} showed an upper bound of
185: 43/16 = 2.6875 for online square packing
186: and an upper bound 6.25 for online cube packing.
187: The upper bound for square packing was improved to
188: 395/162 $<$ 2.43828 by Seiden and van Stee \cite{SS03}.
189: For online cube packing,
190: Miyazawa and Wakabayashi \cite{MW03} showed an upper bound of 3.954.
191: Epstein and van Stee \cite{ES04} gave an upper bound of 2.2697
192: for square packing
193: and an upper bound of 2.9421 for online cube packing.
194: By using a computer program,
195: the upper bound for square packing was improved to 2.24437 by
196: Epstein and van Stee \cite{ES05b}.
197: They \cite{ES05b} also gave lower bounds of 1.6406 and 1.6680
198: for square packing and cube packing, respectively.
199:
200: \paragraph{\bf Our contributions.}
201: When the Harmonic algorithm \cite{LL85} is extended into
202: the online hypercube packing problem,
203: the items of sizes $1/2 + \epsilon, 1/3 + \epsilon, 1/4+\epsilon,
204: \dots$ are still the crucial items related to
205: the asymptotic competitive ratio,
206: where $\epsilon >0$ is sufficiently small.
207: Using the techniques in one dimensional bin packing,
208: Epstein and van Stee \cite{ES05b} combined the items of size in
209: $(1/2, 1-\Delta]$ with the items of size in $(1/3, \Delta]$
210: and improved the Harmonic algorithm for hypercube packing,
211: where $\Delta$ is a specified number in $(1/3, 0.385)$.
212: In this paper, we do not only consider the combinatorial packing
213: for the items in $(1/2, 1-\Delta]$ and $(1/3, \Delta]$,
214: but also other crucial items.
215: Based on the techniques in one dimensional bin packing
216: algorithm Super Harmonic by Seiden \cite{s02},
217: we classify all the items into 17 groups and
218: give a framework for online hypercube packing.
219: To analyse our algorithm,
220: we give a weighting system consisting of four weighting functions.
221: By the weighting functions, we show that
222: for square packing, the asymptotic competitive ratio of our algorithm
223: is at most 2.1439 which is better than 2.24437\cite{ES05b},
224: for cube packing, the ratio is at most 2.6852,
225: which is also better than 2.9421\cite{ES05b}.
226:
227:
228: \noindent {\bf Definition}:
229: If an item $p$ of size (side length) $s(p) \le 1/M$,
230: where $M$ is a fixed integer,
231: then call $p$ {\em small}, otherwise {\em large}.
232:
233:
234:
235: %% \section{Previous Research}
236: %% We first introduce previous research on online hypercube packing
237: %% that is useful in our algorithm.
238: \section{Online packing small items}
239: %% When an itme $h$ of size $s(h) \le 1/M$, where $M \ge 5$ is a fixed integer,
240: %% we call $p$ {\em small}.
241:
242: The following algorithm for packing small items is from
243: \cite{CV93}, \cite{ES05a}.
244: The key ideas are below:
245: \begin{enumerate}
246: \item Classify all {\em small} squares into $M$ groups.
247: In detail, for an item $p$ of size $s(p)$,
248: we classify it into group $i$ such that
249: $2^ks(p) \in (1/(i+1),1/i]$, where $i \in \{M,...,2M-1\}$
250: and $k$ is an integer.
251: \item Exclusively pack items of the same group into bins,
252: i.e., each bin is used to pack items belonged to the same group.
253: During packing, one bin may be partitioned into sub-bins.
254: \end{enumerate}
255:
256:
257: \noindent {\bf Definition}:
258: An item is defined to be of type $i$ if it belongs to group $i$.
259: A sub-bin which received an item is said to be {\em used}.
260: A sub-bin which is not used and not cut into smaller sub-bins is called
261: {\em empty}.
262: A bin is called {\em active} if it can still receive items,
263: otherwise {\em closed}.
264:
265: Given an item $p$ of type $i$, where $2^ks(p) \in (1/(i+1),1/i]$,
266: {\em algorithm} AssignSmall($i$) works as followings.
267:
268: %%\noindent{\bf AssignSmall($i$)}:
269: \begin{enumerate}
270: \item If there is an empty sub-bin of size $1/(2^ki)$,
271: then the item is simply packed there.
272: \item Else, in the current bin, if there is no empty sub-bin
273: of size $1/(2^ji)$ for $j < k$, then close the bin
274: and open a new bin and partition it into sub-bins of size $1/i$.
275: If $k=0$ then pack the item in one of sub-bins of size $1/i$.
276: Else goes to next step.
277: \item Take an empty sub-bin of size $1/(2^ji)$ for a maximum $j < k$.
278: Partition it into $2^d$ identical sub-bins.
279: If the resulting sub-bins are larger than $1/(2^ki)$,
280: then take {\em one} of them and partition it in the same way.
281: This is done until sub-bins of size $1/(2^ki)$ are reached.
282: Then the item is packed into one such sub-bin.
283: \end{enumerate}
284:
285: %% The following results are from \cite{ES05a}.
286: \begin{lemma} \label{lemma:small}
287: In the above algorithm,
288: \vspace{-5pt}
289: \begin{description}
290: \item i) at any time, there are at most $M$ {\em active} bins.
291: \vspace{-5pt}
292: \item ii) in each closed bin of type $i \ge M$,
293: the occupied volume is at least $(i^d -1)/(i+1)^d \ge (M^d -1)/(M+1)^d$.
294: \end{description}
295: \end{lemma}
296: So, roughly speaking, a small item with size $x$ takes
297: at most $\frac{(M+1)^d} {(M^d -1)}\times x^d$ bin.
298:
299: \section{Algorithm $\mathcal{A}$ for online hypercube packing}
300: The key points in our online algorithm are
301: \begin{enumerate}
302: \item divide all items into {\em small} and {\em large} groups.
303: \item pack small items by algorithm AssignSmall,
304: pack large items by an extended Super Harmonic algorithm.
305: \end{enumerate}
306:
307:
308: Before giving our algorithm,
309: we first give some definitions and descriptions about the algorithm,
310: which are similar with the ones in \cite{s02}, but some definitions
311: are different from the ones in \cite{s02}.
312:
313: \noindent{\bf Classification of large items}:
314: Given an integer $M \ge 11$,
315: let $t_1 = 1 > t_2 > \cdots>t_{N+1} = 1/M > t_{N+2}=0$,
316: where $N$ is a fixed integer.
317: We define the interval $I_j$ to be $(t_{j+1},t_j]$ for
318: $j = 1,...,N+1$
319: and say a large item $p$ of size $s(p)$ has type $i$
320: if $s(p) \in I_i$.
321:
322:
323: \noindent{\bf Definition}: An item of size $s$ has type $\tau(s)$, where
324: \[
325: \tau(s) = j \qquad \Leftrightarrow \qquad s \in I_j.
326: \]
327:
328: \noindent{\bf Parameters in algorithm $\mathcal{A}$}:
329: An instance of the algorithm is described by the following parameters:
330: integers $N$ and $K$; real numbers
331: $1 = t_1 > t_2 > \cdots > t_N > t_{N+1} = 1/M$,
332: $\alpha_1,...,\alpha_{N} \in [0,1]$
333: and $0 = \Delta_0 < \Delta_1 < \cdots< \Delta_K< 1/2$,
334: and a function $\phi:\{1,...,N\} \mapsto \{0,...,K\}$.
335:
336: Next, we give the operation of our algorithm,
337: essentially, which is quite similar with
338: the Super Harmonic algorithm \cite{s02}.
339: Each {\em large} item of type $j$ is assigned a color,
340: {\em red} or {\em blue}.
341: The algorithm uses two sets of counters,
342: $e_1,...,e_{N}$ and $s_1,...,s_N$,
343: all of which are initially zero.
344: $s_i$ keeps track of the total number of type $i$ items.
345: $e_i$ is the number of type $i$ items which get colored red.
346: For $1 \le i \le N$,
347: the invariant $e_i = \lfloor \alpha_i s_i \rfloor$ is maintained,
348: i.e. the percentage of type $i$ items colored red is
349: approximately $\alpha_i$.
350:
351:
352: We first introduce some parameters used in Super Harmonic algorithm,
353: then give the corresponding ones for $d$-dimensional packing.
354: In one dimensional packing,
355: a bin can be placed at most
356: $\beta_i = \lfloor 1/t_i\rfloor$ items
357: with size $t_i$.
358: After packing $\beta_i$ type $i$ items,
359: there is
360: $\delta_i = 1 - t_i \beta_i$ space left.
361: The rest space can be used for red items.
362: %% For example, when a type $4$ is placed in a bin,
363: %% in every dimension of a bin, there is a rest space as least
364: %% $\delta_4 = p_1 \ge 1/3$,
365: %% we would like to use this space to pack red items of type $6,7$ etc.
366: However, we sometimes use less than $\delta_i$ in a bin
367: in order to simplify the algorithm and its analysis,
368: i.e.,
369: we use $\mathcal{D}=\{ \Delta_1,...,\Delta_K\}$ instead of the set of
370: $\delta_i$, for all $i$.
371: $\Delta_{\phi(i)}$ is the amount of space used to hold red items in
372: a bin which holds blue items of type $i$.
373: We therefore require that $\phi$ satisfy $\Delta_{\phi(i)} \le \delta_i$.
374: $\phi(i)=0$ indicates that no red items are accepted.
375: To ensure that every red item potentially can be packed,
376: we require that $\alpha_i =0$ for all $i$ such that
377: $t_i > \Delta_K$, that is, there are no red items of type $i$.
378: Define $\gamma_i=0$ if $t_i > \delta_K$ and $\gamma_i = \max\{1, \lfloor \Delta_1/t_i\rfloor\}$, otherwise.
379: This is the number of red item of type $i$ placed in a bin.
380:
381: In $d$-dimensional packing,
382: we place $\beta_i^d$ blue items of type $i$
383: into a bin and introduce a new parameter $\theta_i$ instead of
384: $\gamma_i$.
385: Let
386: \[
387: \theta_i = \beta_i^d
388: -(\beta_i - \gamma_i)^d.
389: \]
390: This is the number of red items of type $i$ that the algorithm
391: places together in a bin.
392: In details,
393: if $t_i > \Delta_K$, then $\theta_i = 0$, i.e., we do not pack
394: type $i$ items as red items.
395: So, in this case, we require $\alpha_i = 0$.
396: Else if $t_i \le \Delta_1$, then $\theta_i = \beta_i^d
397: -(\beta_i - \lfloor \Delta_1/t_i\rfloor)^d$.
398: If $\Delta_1 < t_i \le \Delta_K$, we set
399: $\theta_i = \beta_i^d
400: -(\beta_i - 1)^d$.
401:
402:
403: %% Define $\gamma_i = 0$ if $t_i > \Delta_K$ and
404: %% $\gamma_i = \max\{1, \lfloor \Delta_1/t_i\rfloor\}$ otherwise.
405:
406: %% And if possible, we would like to pack red items into a bin with
407: %% $\beta_i^d$ blue items of type $i$ when $\delta_i \ne 0$.
408:
409: %% Define
410: %% \[
411: %% \varphi(i) = \min\{j | t_i \le \Delta_j, 1 \le j \le K \}.
412: %% \]
413: %% Intuitively, $\varphi(i)$ is the smallest index
414: %% to indicates that a red item of type $i$ can be packed along with
415: %% blue items of type $j$ in a bin,
416: %% where $\varphi(i) \le j \le K $.
417:
418:
419: Here, we illustrate the structure of a bin for $d=2$.
420: \begin{figure}[htbp]
421: \begin{center}
422: \includegraphics[scale=0.7]{./redblue.eps}
423: \caption{ If the bin is a (i, j) or (i, ?) bin,
424: the amount of area for blue items is $(t_i \beta_i)^2$.
425: The amount of area left is $1-(t_i \beta_i)^2$.
426: The amount of this area actually used for red items
427: is $1 - (1-\Delta_{\phi(i)})^2$,
428: where $\Delta_{\phi(i)} \le \delta_i = 1-t_i \beta_i $.}
429: \label{fig:redblue}
430: \end{center}
431: \end{figure}
432:
433: %% We now specify the bin groups. They are named:
434: {\bf Naming bins:} Bins are named as follows:
435: \begin{displaymath}
436: \begin{array}{l}
437: \{i|\phi_i = 0, 1 \le i \le N,\} \\
438:
439: \{(i,?)|\phi_i \ne 0, 1 \le i \le N,\} \\
440:
441: \{(?,j)|\alpha_j \ne 0, 1 \le j \le N,\} \\
442:
443: \{(i,j)|\phi_i \ne 0, \alpha_j \ne 0, \gamma_j t_j \le \Delta_{\phi(i)},
444: 1 \le i,j \le N\}.
445: \end{array}
446: \end{displaymath}
447: We call these groups {\em monochromatic}, {\em indeterminate blue},
448: {\em indeterminate red} and {\em bichromatic}, respectively.
449: And we call the monochromatic and bichromatic groups {\em final} groups.
450:
451: The monochromatic group $i$ contains bins that hold only blue items of type
452: $i$. There is only one open bin in each of these groups;
453: this bin has fewer than $\beta_i^d$ items.
454: The closed bins all contain $\beta_i^d$ items.
455:
456: The bichromatic group $(i,j)$ contains bins that contain blue items of type
457: $i$ along with red items of type $j$. A closed bin in this group contains
458: $\beta_i^d$ type $i$ items and $\theta_j$ type $j$ items. There are at most
459: three open bins.
460:
461: The indeterminate blue group $(i, ?)$ contains bins that hold only blue items
462: of type $i$. These bins are all open, but only one has fewer than
463: $\beta_i^d$ items.
464:
465: The indeterminate red group $( ?, j)$ contains bins that hold only red items
466: of type $j$. These bins are all open, but only one has fewer than
467: $\theta_j$ items.
468:
469: Essentially, the algorithm tries to minimize the number of indeterminate bins,
470: while maintaining all the aforementioned invariants.
471: That is, we try to place red and blue items together whenever possible;
472: when this is not possible we place them in indeterminate bins in hope that
473: they can later be so combined.
474:
475:
476: %%\baselineskip 11pt
477: \noindent{\bf Algorithm $\mathcal{A}$}:
478: A formal description of algorithm $\mathcal{A}$ is given as blow:
479: \begin{description}
480: \item Initialize $e_i \gets 0$ and $s_i \gets 0$ for $1 \le i \le M+1$.
481: \item For a small item $p$, call algorithm AssignSmall.
482: \item For a large item $p$:
483: \begin{description}
484: \item $i \gets \tau(p)$, \quad $s_i \gets s_i +1$.
485: \item If $e_i < \lfloor \alpha_i s_i \rfloor$:
486: \begin{description}
487: \item $e_i \gets e_i +1$.
488: \item Color $p$ red.
489: \item If there is an open bin in group $(?,i)$ with
490: fewer than $\theta_i$ type $i$ items, then pack $p$ in
491: this bin.
492: \item If there is an open bin in group $(j,i)$ with
493: fewer than $\theta_i$ type $i$ items, then pack $p$ in
494: this bin.
495: \item Else if there is some bin in group $(j,?)$ such that
496: $\Delta_{\phi(j)} \ge \gamma_i t_i$ then place
497: $p$ in it and change the group of this bin to $(j,i)$.
498: \item Otherwise, open a new group $(?, i)$ bin and place $p$ in it.
499: \end{description}
500: \item Else:
501: \begin{description}
502: \item Color $p$ blue.
503: \item If $\phi(i) = 0 $:
504: \begin{description}
505: \item If there is an open bin in group $i$ with fewer than $\beta_i^d$
506: items, then place $p$ in it.
507: \item Otherwise, open a new group $i$ bin and pack $p$ there.
508: \end{description}
509: \item Else:
510: \begin{description}
511: \item If, for any $j$, there is an open bin $(i,j)$ with fewer
512: than $\beta_i^d$ items, then place $p$ in this bin.
513:
514: \item Else, if there is some bin in group $(i, ?)$ with fewer
515: than $\beta_i^d$ items, then place $p$ in this bin.
516: \item Else, if there is some bin in group $(?,j)$ such that
517: $\Delta_{\phi(i)} \ge \gamma_j t_j$ then pack $p$ in it
518: and change the group of this bin to $(i,j)$.
519: \item Otherwise, open a new group $(i,?)$ bin and pack $p$ there.
520: \end{description}
521: \end{description}
522: \end{description}
523: \end{description}
524:
525: \baselineskip 13.4pt
526:
527: \section{The analyses for square and cube packing }
528: In this section, we fix the parameters in the framework given
529: in the last section for square packing and cube packing respectively.
530: Then we analyse the competitive ratios by
531: a corresponding weighting system consisting
532: of four weighting functions.
533:
534:
535: \subsection{An instance of algorithm $\mathcal{A}$}
536: Let $M=11$, i.e., a {\em small} item has its side length as most $1/11$.
537: And the parameters in $\mathcal{A}$ are given in the following tables.
538: First we classify all the items into 17 groups by fixing the values of
539: $t_i$, where $1 \le i \le 18$.
540: Then we calculate the number of blue type $i$ in a bin, $\beta_i^{d}$.
541: Finally, we define the set $\mathcal{D}=\{ \Delta_1,...,\Delta_K\}$
542: and the function $\phi(i)$,
543: which are related to how many red items $\theta_i^{d}$ can be accepted
544: in a bin, where $K=4$.
545: Note that $\alpha_i$ which is the percentage of type $i$ items colored red
546: will be given later. For square packing, we use a set of $\alpha_i$.
547: While for cube packing, we use another set of $\alpha_i$.
548:
549:
550: %%\begin{table}[htbp]
551: \begin{displaymath}
552: \begin{array}{|c|c|c|c|c|c|}
553: \hline
554: i & (t_{i+1},t_i] & \beta_i & \delta_i & \phi(i) & \gamma_i \\
555: \hline
556: 1 & (0.7, 1] & 1 & 0 & 0 & 0 \\
557: %% \hline
558: 2 & (0.65, 0.7] & 1 & 0.3 & 2 & 0 \\
559: %% \hline
560: 3 & (0.60, 0.65] & 1 & 0.35 & 3 & 0 \\
561: %% \hline
562: 4 & (0.5, 0.60] & 1 & 0.4 & 4 & 0 \\
563: %% \hline
564: 5 & (0.4, 0.5 ] & 2 & 0 & 0 & 0 \\
565: %% \hline
566: 6 & (0.35, 0.4] & 2 & 0.2 & 1 & 1 \\
567: %% \hline
568: 7 & (1/3, 0.35] & 2 & 0.3 & 2 & 1 \\
569: %% \hline
570: 8 & (0.30, 1/3] & 3 & 0 & 0 & 0 \\
571: %%\hline
572: 9 & (1/4, 0.30] & 3 & 0.1 & 0 & 1 \\
573: %%\hline
574: 10 & (1/5, 1/4 ] & 4 & 0 & 0 & 1 \\
575: %%%\hline
576: 11 & (1/6, 1/5 ] & 5 & 0 & 0 & 1 \\
577: %%\hline
578: 12 & (1/7, 1/6] & 6 & 0 & 0 & 1 \\
579: 13 & (1/8, 1/7] & 7 & 0 & 0 & 1 \\
580: 14 & (1/9, 1/8] & 8 & 0 & 0 & 1 \\
581: 15 & (0.1, 1/9] & 9 & 0 & 0 & 1 \\
582: 16 & (1/11, 0.1] & 10 & 0 & 0 & 2 \\
583: 17 & (0, 1/11] & * & * & * & * \\
584: \hline
585: \end{array}
586: \hspace{3mm}
587: \begin{array}{|c|c|c|}
588: \hline
589: j = \phi(i) & \Delta_j & \textrm{Red items accepted} \\
590: \hline
591: 1 & 0.20 & 11..16 \\
592: 2 & 0.30 & 9..16 \\
593: 3 & 0.35 & 7, 9..16 \\
594: 4 & 0.40 & 6..7, 9..16 \\
595:
596: \hline
597: \end{array}
598: \end{displaymath}
599:
600:
601: \noindent{\bf Observation:} By the above tables,
602: in any dimension of a $(4,?)$ bin,
603: the distance between the type 4 item and the opposite edge (face) of
604: the bin is at least $\Delta_4 = 0.4$,
605: since we pack a type 4 item in a corner of a bin.
606: So, all red items with size at most 0.4 can be packed in $(4,?)$ bins.
607: In the same ways,
608: all red items with size at most 0.35 can be packed in $(4,?)$ and $(3,?)$ bins,
609: all red items with size at most 0.30 can be packed in
610: $(4,?)$, $(3,?)$, $(7,?)$ and $(2,?)$ bins,
611: all red items with size at most 0.2 can be packed in
612: $(4,?)$, $(3,?)$, $(7,?)$, $(2,?)$, $(6,?)$ bins.
613:
614: Next we define the weight function $W(p)$ for a given item $p$ with size $x$.
615: Roughly speaking, a weight of an item is the maximal portion of a bin
616: that it can occupy.
617: Given a small item $p$ with size $x$, by Lemma \ref{lemma:small},
618: it occupies a $\frac{x^d (11+1)^d}{11^d -1}$ bin.
619: So, we define
620: \[
621: W(p) = \frac{x^d (11+1)^d}{11^d -1}.
622: \]
623:
624: Given a large item $p$, we consider four cases to
625: define its weight.
626: Let $R_i$ and $B_i$ be the number of bins containing blue items
627: of type $i$ and red items of type $i$, respectively.
628: Let $E$ be the number of indeterminate red group bins,
629: i.e., some bins like $(?,i)$.
630: If $E > 0$ then there are some $(?, j)$ bins.
631: Let
632: \[
633: e = \min\{j| (?,j)\},
634: \]
635: which is the type of
636: the smallest red item in an indeterminate red group bin.
637: Let $\mathcal{A}(L)$ be the number of bins used by $\mathcal{A}$.
638:
639: \noindent {\bf Case 1}: $E=0$, i.e., no indeterminate red bins.
640: Then every red item is packed with one or more blue items.
641: Therefore
642: \[
643: \mathcal{A}(L) \le \mathcal{A}(L_s) + \sum_{i} B_i,
644: \]
645: where $\mathcal{A}(L_s)$ is the number of bins for small items.
646: Since there are a constant number of active bins and
647: every closed blue bin $(i)$ or $(i,*)$ contains $\frac{1}{\beta_i^d}$ items,
648: we define the weighting function as below:
649: \begin{displaymath}
650: W_{1,1}(p) = \frac{1-\alpha_i}{\beta_i^d} \qquad \textrm{if } x \in I_i, \textrm{ for } i=1..16.
651: \end{displaymath}
652:
653: %% \begin{displaymath}
654: %% \begin{array}{ll}
655: %% W_1(p) = & \left\{
656: %% %% { \setlength \arrayrowsep{1mm}
657: %% \begin{array}{ll}
658: %% \frac{1}{\beta_i^d} & \textrm{if } x \in I_i, \textrm{ for } i=1..5,8. \\
659: %% \vspace{4pt}
660: %% \frac{1-\alpha_i}{\beta_i^d} & \textrm{if } x \in I_i, \textrm{ for } i=6,7,9..16.
661: %% \end{array}
662: %% \right.
663: %% \end{array}
664: %% \end{displaymath}
665:
666:
667:
668: \noindent {\bf Case 2}: $E>0$ and $e = 6$. Then
669: there are some bins $(?,6)$ and no other bins $(?,j)$ bins,
670: where $j > 6$. Since a type 4 item can be packed
671: into a bin $(?,6)$, it is impossible to have bins $(4, ?)$.
672: If we count all $(4,j)$ bins as red bins,
673: then
674: \[
675: \mathcal{A}(L) \le \mathcal{A}(L_s) + \sum_{i=1..3,5,8} B_i
676: + \sum_{i=6,7,9..16} (R_i + B_i).
677: \]
678: Else we count all $(4,j)$ bins as blue bins then
679:
680: \[
681: \mathcal{A}(L) \le \mathcal{A}(L_s) + \sum_{i=1..16} B_i
682: + R_6.
683: \]
684: Since there are a constant number of active bins and
685: every closed blue bin $(i)$ or $(i,*)$ contains $\frac{1}{\beta_i^d}$ items,
686: every closed red bin $(j,i)$ or $(?,i)$ contains $\frac{1}{\theta_i}$ items,
687: we define the weighting functions for two subcases as below:
688: \begin{displaymath}
689: \begin{array}{ll}
690: W_{2,1}(p) = & \left\{
691: \begin{array}{ll}
692: \frac{1}{\beta_i^d} & \textrm{if } x \in I_i,
693: \textrm{ for } i= 1,2,3,5,8. \\
694: \vspace{4pt}
695: 0 & \textrm{if } x \in I_4. \\
696: %% \vspace{4pt}
697: %% \frac{1-\alpha_i}{\beta_i^d} + \frac{\alpha_i}{\theta_i} & \textrm{if } x \in I_i, \textrm{ for } i= 6.\\
698: \vspace{4pt}
699: \frac{1-\alpha_i}{\beta_i^d} + \frac{\alpha_i}{\theta_i} & \textrm{if } x \in I_i, \textrm{ for } i=6,7,9..16.
700: \end{array}
701: \right.
702: \end{array}
703: \end{displaymath}
704: and
705: \begin{displaymath}
706: \begin{array}{ll}
707: W_{2,2}(p) = & \left\{
708: \begin{array}{ll}
709: \frac{1-\alpha_i}{\beta_i^d} & \textrm{if } x \in I_i,
710: \textrm{ for } i= 1..5,7..16. \\
711: %% \vspace{4pt}
712: %% 1 & \textrm{if } x \in I_4. \\
713: \vspace{4pt}
714: \frac{1-\alpha_i}{\beta_i^d} + \frac{\alpha_i}{\theta_i} & \textrm{if } x \in I_i, \textrm{ for } i= 6.
715: %% \vspace{4pt}
716: %% \frac{1-\alpha_i}{\beta_i^d} & \textrm{if } x \in I_i, \textrm{ for } i=7,9..16.
717: \end{array}
718: \right.
719: \end{array}
720: \end{displaymath}
721:
722:
723: \noindent {\bf Case 3}: $E>0$ and $e = 7$. Then
724: there are some bins $(?,7)$ and no other bins $(?,j)$,
725: where $j > 7$. Since a type 4 or a type 3 item can be packed
726: into a bin $(?,7)$, it is impossible to have bins $(4, ?)$ and $(3, ?)$.
727: If we count all $(4,j)$ and $(3, j)$ bins as red bins,
728: then
729: \[
730: \mathcal{A}(L) \le \mathcal{A}(L_s) + \sum_{i=1,2,5,8} B_i
731: + \sum_{i=6,7,9..16} (R_i + B_i).
732: \]
733: Else we count all $(4,j)$ and $(3, j)$ bins as blue bins then
734:
735: \[
736: \mathcal{A}(L) \le \mathcal{A}(L_s) + \sum_{i=1..16} B_i
737: + R_6 + R_7.
738: \]
739: %% Since there are a constant number of active bins and
740: %% every closed blue bin $(i)$ or $(i,*)$ contains $\frac{1}{\beta_i^d}$ items,
741: %% every closed red bin $(j,i)$ or $(?,i)$ contains $\frac{1}{\theta_i}$ items,
742: We define the weighting functions for two subcases as below:
743: \begin{displaymath}
744: \begin{array}{ll}
745: W_{3,1}(p) = & \left\{
746: \begin{array}{ll}
747: \frac{1}{\beta_i^d} & \textrm{if } x \in I_i,
748: \textrm{ for } i= 1,2,5,8. \\
749: \vspace{4pt}
750: 0 & \textrm{if } x \in I_3, I_4. \\
751: %% \vspace{4pt}
752: %% \frac{1-\alpha_i}{\beta_i^d} + \frac{\alpha_i}{\theta_i} & \textrm{if } x \in I_i, \textrm{ for } i= 6.\\
753: \vspace{4pt}
754: \frac{1-\alpha_i}{\beta_i^d} + \frac{\alpha_i}{\theta_i} & \textrm{if } x \in I_i, \textrm{ for } i=6,7,9..16.
755: \end{array}
756: \right.
757: \end{array}
758: \end{displaymath}
759: and
760: \begin{displaymath}
761: \begin{array}{ll}
762: W_{3,2}(p) = & \left\{
763: \begin{array}{ll}
764: \frac{1-\alpha_i}{\beta_i^d} & \textrm{if } x \in I_i,
765: \textrm{ for } i= 1..5,8..16. \\
766: %% \vspace{4pt}
767: %% 1 & \textrm{if } x \in I_4. \\
768: \vspace{4pt}
769: \frac{1-\alpha_i}{\beta_i^d} + \frac{\alpha_i}{\theta_i} & \textrm{if } x \in I_i, \textrm{ for } i= 6, 7.
770: %% \vspace{4pt}
771: %% \frac{1-\alpha_i}{\beta_i^d} & \textrm{if } x \in I_i, \textrm{ for } i=7,9..16.
772: \end{array}
773: \right.
774: \end{array}
775: \end{displaymath}
776:
777: %% \begin{displaymath}
778: %% \begin{array}{ll}
779: %% W_{3,1}(p) = & \left\{
780: %% \begin{array}{ll}
781: %% \frac{1}{\beta_i^d} & \textrm{if } x \in I_i, \textrm{ for } i=1,2,5,8. \\
782: %% \vspace{4pt}
783: %% 0 & \textrm{if } x \in I_i, \textrm{ for } i=3,4. \\
784: %% \vspace{4pt}
785: %% \frac{1-\alpha_i}{\beta_i^d} + \frac{\alpha_i}{\theta_i} & \textrm{if } x \in I_i, \textrm{ for } i=6,7.\\
786: %% \vspace{4pt}
787: %% \frac{1-\alpha_i}{\beta_i^d} + \frac{\alpha_i}{\theta_i} & \textrm{if } x \in I_i, \textrm{ for } i=9..16.
788: %% \end{array}
789: %% \right.
790: %% \end{array}
791: %% \end{displaymath}
792: %% and
793: %% \begin{displaymath}
794: %% \begin{array}{ll}
795: %% W_{3,2}(p) = & \left\{
796: %% \begin{array}{ll}
797: %% \frac{1}{\beta_i^d} & \textrm{if } x \in I_i, \textrm{ for } i=1,2,5,8. \\
798: %% \vspace{4pt}
799: %% 1 & \textrm{if } x \in I_i, \textrm{ for } i=3,4. \\
800: %% \vspace{4pt}
801: %% \frac{1-\alpha_i}{\beta_i^d} + \frac{\alpha_i}{\theta_i} & \textrm{if } x \in I_i, \textrm{ for } i=6,7.\\
802: %% \vspace{4pt}
803: %% \frac{1-\alpha_i}{\beta_i^d} & \textrm{if } x \in I_i, \textrm{ for } i=9..16.
804: %% \end{array}
805: %% \right.
806: %% \end{array}
807: %% \end{displaymath}
808:
809:
810:
811: \noindent {\bf Case 4}: $E>0$ and $e \ge 9$. Then
812: there are some bins $(?,9)$. Since a type 2,3,4,7 item can be packed
813: into a bin $(?,9)$, it is impossible to have bins $(2, ?)$, $(3, ?)$,
814: $(4, ?)$, $(7, ?)$.
815: If we count these bins $(2, j)$, $(3, j)$,
816: $(4, j)$, $(7, j)$ as red bins,
817: then
818: \[
819: \mathcal{A}(L) \le \mathcal{A}(L_s) + \sum_{i=1,5,8} B_i
820: + \sum_{i=6,9..16} (R_i + B_i) + R_7.
821: \]
822: %% Else we count all $(4,j)$ and $(3, j)$ bins as blue bins then
823:
824: %% \[
825: %% \mathcal{A}(L) \le \mathcal{A}(L_s) + \sum_{i=1..16} B_i
826: %% + R_6 + R_7.
827: %% \]
828: We define the weighting function as below:
829: \begin{displaymath}
830: \begin{array}{ll}
831: W_{4,1}(p) = & \left\{
832: \begin{array}{ll}
833: \frac{1}{\beta_i^d} & \textrm{if } x \in I_i, \textrm{ for } i=1,5,8. \\
834: \vspace{4pt}
835: 0 & \textrm{if } x \in I_2,I_3, I_4 \\
836: \vspace{4pt}
837: \frac{1-\alpha_i}{\beta_i^d} + \frac{\alpha_i}{\theta_i} & \textrm{if } x \in I_i, \textrm{ for } i=6,9..16 \\
838: \vspace{4pt}
839: \frac{\alpha_i}{\theta_i} & \textrm{if } x \in I_i, \textrm{ for } i=7.
840: %% \vspace{4pt}
841: %% \frac{1-\alpha_i}{\beta_i^d} & \textrm{if } x \in I_i, \textrm{ for } i=11..16.
842: \end{array}
843: \right.
844: \end{array}
845: \end{displaymath}
846:
847:
848:
849:
850:
851: \noindent {\bf Definition}:
852: A set of items $X$ is a feasible set
853: if all items in it can be packed into a bin.
854: And,
855: \[
856: W_{i,j}(X) = \sum_{p \in X} W_{i,j}(p) .
857: \]
858:
859:
860: Over all feasible sets $X$, let
861: \[
862: W_i(X) = \min\{W_{i,j}(X)\}, \textrm { j = 1 or 2},
863: \]
864: and define
865: \[
866: \mathcal{P}(W) = \max\{ W_i(X)\} \textrm { for all } i.
867: \]
868:
869:
870: We defined four sets of weighting functions for all items.
871: This is a weighting system,
872: which is a special case of general weighting system defined in \cite{s02}.
873: So, the following lemma follows directly from \cite{s02}.
874: \begin{lemma}
875: \label{lemma:upper}
876: The asymptotic performance ratio of $\mathcal{A}$ is upper bounded by $\mathcal{P}(W)$.
877: \end{lemma}
878:
879:
880: \subsection{Upper bounds for square and cube packing}
881: In this subsection,
882: we fix the parameters $\alpha_i$ for square packing
883: and cube packing respectively,
884: and get the upper bounds of the asymptotic competitive ratios.
885:
886: \noindent{\bf Definition} Let $m_i \ge 0$ be the number of
887: type $i$ items in a feasible set $X$.
888: Given an item $p$ with size $x$,
889: define an efficient function $E_{i,j}(p)$ as $W_{i,j}(p)/x^d$.
890:
891: \begin{theorem}
892: The asymptotic performance ratio of $\mathcal{A}$ for square packing is at most 2.1439.
893: \end{theorem}
894: \begin{proof}
895: For square packing, we set parameters $\alpha_i$ according to
896: the following table.
897: \begin{displaymath}
898: \begin{array}{|c|l|l|l|l|l|l|l|l|l|l|l|l|l|}
899: \hline
900: i & 1-4 & 5 & 6 & 7 & 8 & 9 & 10 & 11 & 12& 13& 14 &15 & 16 \\
901: \hline
902: \alpha_i & 0 & 0& 0.12 & 0.2 & 0 & 0.2546 & 0.2096 & 0.15& 0.1& 0.1 &0.1& 0.1 & 0.05 \\
903: \hline
904: \theta_i & 0 &0 & 3 & 3 & 0 & 5 & 7 & 9 &11& 13 & 15 & 17 & 36 \\
905: \hline
906: \beta_i^2& 1 & 4 & 4 &4 & 9& 9 & 16& 25 & 36& 49& 64 & 81 & 100 \\
907: \hline
908: \end{array}
909: \end{displaymath}
910:
911:
912:
913: Based on the values in the followint two tables,
914: we calculate the upper bound of $\mathcal{P}(W) = \max\{W_i(X)\}$.
915: \begin{displaymath}
916: \begin{array}{|c|l|l|l|l|l|l|l|}
917: \hline
918: i & (t_{i+1},t_i] & W_{1,1}(p) & E_{1,1}(p) & W_{2,1}(p) & E_{2,1}(p) &W_{2,2}(p) & E_{2,2}(p) \\
919: \hline
920: 1 &(0.7,1] & 1 & 2.05 & 1 & 2.05 & 1 & 2.05 \\
921: 2 &(0.65,0.7]& 1 & 2.37 & 1 & 2.37 & 1 & 2.37 \\
922: 3 &(0.6,0.65]& 1 & 2.7778 & 1 & 2.7778& 1 & 2.7778 \\
923: 4 &(0.5,0.6] & 1 & 4 & 0 & 0 & 1 & 4 \\
924: 5 &(0.4,0.5]& 1/4 & 1.5625 & 1/4 & 1.5625 & 1/4 & 1.5625 \\
925: 6 & (0.35,0.4]& 0.22 & 1.8 & 0.26 & 2.123 & 0.26 & 2.123 \\
926: 7 &(1/3,0.35]& 0.2 & 1.8 & 0.8/3 & 2.4 & 0.2 & 1.8 \\
927: 8 &(0.3,1/3]& 1/9 & 1.235 & 1/9 & 1.235 & 1/9 & 1.235 \\
928: 9 &(1/4,0.3]& 0.0829 & 1.327 & 0.1338 & 2.141 & 0.0829 & 1.327 \\
929: 10..17 &(0,1/4]& 1.235x^2& 1.235 &1.99x^2 & 1.99 &1.235x^2 & 1.235 \\
930: %% 11 &(1/6,1/5]& 0.034& 1.224 & 0.05067 & 1.824 & 0.034 & 1.224 \\
931: %% 12 &(1/7,1/6]& 0.025 & 1.225 & 0.03410 & 1.6705& 0.025 & 1.225 \\
932: %% 13 &(1/8,1/7]& 0.01837 & 1.1756 & 0.02606 & 1.6679 & 0.01837 & 1.1756 \\
933: %% 14 &(1/9,1/8]& 0.9/64 &1.2 & 0.02073 & 1.6791 & 0.9/64 &1.2 \\
934: %% 15 &(0.1,1/9]& 0.1/9& 1.2 & 0.017 & 1.7 & 0.1/9 & 1.2 \\
935: %% 16 &(1/11,0.1]& 0.0095 & 1.2 & 0.01089 & 1.3176 & 0.0095 & 1.2 \\
936: %% 17 &(0,1/11]& 1.2x^2 & 1.2 &1.2x^2 & 1.2 &1.2x^2 & 1.2 \\
937: \hline
938: \end{array}
939: \end{displaymath}
940:
941: %% Based on the number of items $m_i$.
942: %% we have 5 cases for
943: %% estimating the upper bound of $\mathcal{P}(W) = \max\{W_i(X)\}$.
944:
945:
946: \paragraph{ Case 1:} $ W_1(X) \le 2.1439$.
947:
948: \noindent If $m_2 + m_3 + m_4 = 0$, i.e., no type $2,3,4$ items in $X$,
949: then
950: \[
951: W_1(X) = \sum_{p \in X} E_{1,1}(p) s(p)^2
952: \le 2.05 \sum_{p \in X} s(p)^2 \le 2.05.
953: \]
954: Else $m_2 + m_3 + m_4 = 1$. Then
955: $m_5 + m_6 + m_7 \le 3$ and $m_6 + m_7 + m_9 \le 5$,
956: \[
957: \begin{array}{lll}
958: W_1(X) & \le & 1+ m_5/4 + 0.22 m_6 + 0.2m_7 + 0.0829m_9
959: + 1.235(1- \sum_{i=2}^{7}t_{i+1}^2 m_i -m_9/16) \\
960: & < & 2.1439.
961: \end{array}
962: \]
963: The last inequality follows from $m_4=1$, $m_6=3$ and $m_9=2$.
964:
965: \paragraph{ Case 2:} $ W_2(X) \le 2.134$.
966:
967: \noindent If $m_2 + m_3 + m_4 = 0$, i.e., no type $2,3,4$ items in $X$,
968: then
969: \[
970: W_2(X) = \min\{ W_{2,1}(X), W_{2,2}(X)\} \le W_{2,2}(X) \le 2.123.
971: \]
972: Else $m_2 = 1$. Then no type $1,3,4,5,6$ items in $X$.
973: \[
974: W_2(X) = W_{2,2}(X) \le 1 + 1.8 (1 - 0.65^2) = 2.0395.
975: \]
976: Else $m_3 = 1$. Then no type $1,2,4,5$ items in $X$ and
977: $ m_6 + m_7 \le 3$ and $m_6 + m_7 + m_9 \le 5$,
978: \[
979: \begin{array}{lll}
980: W_2(X) = W_{2,2}(X) &\le& 1+ 0.26 m_6 + 0.2m_7 + 0.0829m_9 \\
981: & & + 1.235(1-0.6^2 -0.35^2m_6-m_7/9 -m_9/16) \\
982: & < & 2.134.
983: \end{array}
984: \]
985: The last inequality follows from $m_6=3$ and $m_9=2$.
986:
987: \noindent Else $m_4 =1$. Then no type $1,2,3$ items in $X$.
988: \[
989: W_2(X) \le W_{2,1}(X) \le 0 + 2.4 (1 - 0.5^2) = 1.8.
990: \]
991:
992: \begin{displaymath}
993: \begin{array}{|c|l|l|l|l|l|l|l|}
994: \hline
995: i & (t_{i+1},t_i] & W_{3,1}(p) & E_{3,1}(p) & W_{3,2}(p) & E_{3,2}(p) & W_{4,1}(p) & E_{4,1}(p) \\
996: \hline
997: 1 &(0.7,1] & 1 & 2.05 & 1 & 2.05 & 1 & 2.05 \\
998: 2 &(0.65,0.7] & 1 & 2.37 & 1 & 2.37 & 0 & 0 \\
999: 3 &(0.6,0.65] & 0 & 0 & 1 & 2.7778 & 0 & 0 \\
1000: 4 &(0.5,0.6] & 0 & 0 & 1 & 4 & 0 & 0 \\
1001: 5 &(0.4,0.5] & 1/4 & 1.5625 & 1/4 & 1.5625 & 1/4 & 1.5625\\
1002: 6 & (0.35,0.4] & 0.26 & 2.123 & 0.26 & 2.123 & 0.26 & 2.123 \\
1003: 7 &(1/3,0.35] & 0.8/3 & 2.4 & 0.8/3 & 2.4 & 0.2/3 & 0.6 \\
1004: 8 &(0.3,1/3] & 1/9 & 1.235 & 1/9 & 1.235 & 1/9 & 1.235\\
1005: 9 &(1/4,0.3] & 0.1338 & 2.141 & 0.0829 & 1.327 & 0.1338 & 2.141 \\
1006: 10..17 &(1/5,1/4] &1.99x^2 & 1.99 &1.235x^2& 1.235 &1.99x^2 & 1.99 \\
1007: %% 11 &(1/6,1/5] & 0.05067 & 1.824 & 0.034 & 1.224 & 0.05067 & 1.824 \\
1008: %% 12 &(1/7,1/6] &0.03410 & 1.6705 & 0.025 & 1.225 &0.03410 & 1.6705 \\
1009: %% 13 &(1/8,1/7] & 0.02606 & 1.6679 & 0.01837 & 1.1756 & 0.02606 & 1.6679 \\
1010: %% 14 &(1/9,1/8] & 0.02073 & 1.6791 & 0.9/64 &1.2& 0.02073 & 1.6791 \\
1011: %% 15 &(0.1,1/9] & 0.017 & 1.7 & 0.1/9 & 1.2 & 0.017 & 1.7 \\
1012: %% 16 &(1/11,0.1] & 0.01089 & 1.3176 & 0.0095 & 1.2 & 0.01089 & 1.3176 \\
1013: %% 17 &(0,1/11] &1.2x^2 & 1.2 &1.2x^2 & 1.2 &1.2x^2 & 1.2 \\
1014: \hline
1015: \end{array}
1016: \end{displaymath}
1017:
1018: \paragraph{ Case 3:} $ W_3(X) \le 2.12$.
1019:
1020: \noindent If $m_1+ m_2 + m_3 + m_4 = 0$, i.e.,
1021: no type $1,2,3,4$ items in $X$,
1022: then $m_6 + m_7 \le 4$,
1023: \[
1024: W_3(X) = W_{3,2}(X) \le 0.26m_6 + \frac{0.8m_7}{3} +
1025: 1.5625(1-0.35^2m_6 -\frac{m_7}{9})
1026: < 2.
1027: \]
1028: Else $m_1 =1$ then $m_i=0$, where $2 \le i \le 8$,
1029: \[
1030: W_3(X) = W_{3,2}(X) \le 2.05.
1031: \]
1032: Else $m_2 = 1$. Then no type $1,3,4,5,6$ items in $X$,
1033: $m_7 + m_9 \le 5$ and $m_7 \le 3$.
1034: \[
1035: W_3(X) = W_{3,2}(X) \le 1 + \frac{0.8m_7}{3} + 0.0829m_9 + 1.235(1-0.65^2 - m_7/9 -m_9/16) < 2.12.
1036: \]
1037: Else $m_3 + m_4 = 1$.
1038: Then no type $1,2,3$ items in $X$.
1039: \[
1040: W_3(X) \le W_{3,1}(X) \le 0 + 2.4 (1 - 0.5^2) = 1.8.
1041: \]
1042:
1043: \paragraph{ Case 4:} $ W_4(X) = \sum_{p \in X} E_{4,1}(p) s(p)^2
1044: \le 2.141 \sum_{p \in X} s(p)^2 \le 2.141$.
1045:
1046: %% \noindent {\bf Case 1}:
1047: %% There are no type $1,2,3,4$ items in $X$,
1048: %% i.e., $m_i=0$ for $1 \le i \le 4$.
1049: %% Then,
1050: %% \[
1051: %% W_4(X) = \sum_{p \in X} E_4(p) s(p)^2
1052: %% \le 2.141 \sum_{p \in X} s(p)^2 \le 2.141.
1053: %% \]
1054: %% On the other side,
1055: %% we have $m_6 + m_7 \le 4$ and $W_1(X) \le W_2(X) \le W_3(X)$,
1056: %% \[
1057: %% W_3(X) = \sum_{p \in X} W_3(p) \le 0.26m_6 + \frac{0.8m_7}{3} +
1058: %% 1.5625(1-0.35^2m_6 -\frac{m_7}{9})
1059: %% < 2.
1060: %% \]
1061: %% So, $\mathcal{P}(W) \le \max\{W_i(X)\} \le 2.141$.
1062:
1063: %% \noindent {\bf Case 2}: $m_1 =1$,
1064: %% there is one type $1$ item in $X$. So, $m_i=0$, where $2 \le i \le 8$.
1065: %% Then
1066: %% \[ \mathcal{P}(W) \le 2.141.\]
1067:
1068: %% \noindent {\bf Case 3}: $m_2 =1$, one type $2$ item in $X$.
1069: %% So no type $1,3,4,5,6$ items in $X$.
1070: %% Then we have $ W_4(X) \le 2.141$.
1071: %% And $W_1(X) \le W_2(X) \le W_3(X)$.
1072: %% Since no type $5,6$ items in $X$ and $m_7 + m_9 \le 5$,
1073: %% $m_7 \le 3$, then
1074: %% \[
1075: %% W_3(X) \le 1 + \frac{0.8m_7}{3} + 0.0829m_9 + 1.235(1-0.65^2 - m_7/9 -m_9/16)
1076: %% < 2.12.
1077: %% \]
1078: %% The last inequality follows from $m_7=3$ and $m_9=2$. So,
1079: %% $\mathcal{P}(W) \le 2.141$.
1080:
1081: %% \noindent {\bf Case 4}: $m_3 =1$, one type $3$ item in $X$.
1082: %% So no type $1,2,4,5$ items in $X$.
1083: %% Then
1084: %% \[
1085: %% \max\{W_3(X), W_4(X)\} \le 0 + 2.4(1-0.6^2) < 1.6.
1086: %% \]
1087: %% Else, $W_1(X) \le W_2(X)$, $m_6 + m_7 \le 3$ and
1088: %% $m_6 + m_7 + m_9 \le 5$,
1089: %% \[
1090: %% \begin{array}{lll}
1091: %% W_2(X) &\le& 1+ 0.26 m_6 + 0.2m_7 + 0.0829m_9 \\
1092: %% & & + 1.235(1-0.6^2 -0.35^2m_6-m_7/9 -m_9/16) \\
1093: %% & < & 2.134.
1094: %% \end{array}
1095: %% \]
1096: %% The last inequality follows from $m_6=3$ and $m_9=2$.
1097: %% So,
1098: %% $\mathcal{P}(W) \le 2.134$.
1099:
1100: %% \noindent {\bf Case 5}: $m_4 =1$, one type $4$ item in $X$.
1101: %% So no type $1,2,3$ items in $X$.
1102: %% Then
1103: %% \[
1104: %% \max\{W_2(X), W_3(X), W_4(X)\}\le 0 + 2.4(1-0.5^2) \le 1.8.
1105: %% \]
1106: %% Else $m_5 + m_6 + m_7 \le 3$ and $m_6 + m_7 + m_9 \le 5$,
1107: %% \[
1108: %% \begin{array}{lll}
1109: %% W_1(X) & \le & 1+ m_5/4 + 0.22 m_6 + 0.2m_7 + 0.0829m_9 \\
1110: %% & & + 1.235(1-0.5^2 - 0.4^2m_5 -0.35^2m_6-m_7/9 -m_9/16) \\
1111: %% & < & 2.1439.
1112: %% \end{array}
1113: %% \]
1114: %% The last inequality follows from $m_6=3$ and $m_9=2$.
1115: So, $\mathcal{P}(W) \le 2.1439$.
1116: \end{proof}
1117:
1118: \begin{theorem}
1119: The asymptotic performance ratio of $\mathcal{A}$ for cube packing is at most 2.6852.
1120: \end{theorem}
1121: \begin{proof}
1122: For cube packing, we set parameters $\alpha_i$ and $\theta_i$ in the following table.
1123: \begin{displaymath}
1124: \begin{array}{|c|l|l|l|l|l|l|l|l|l|l|}
1125: \hline
1126: i & 1-4 & 5 & 6 & 7 & 8 & 9 & 10 & 11 & 12-16 \\
1127: \hline
1128: \alpha_i & 0 & 0 & 0.12 & 0.2 & 0 & 0.325 & 0.2096 & 0.15& 0 \\
1129: \hline
1130: \theta_i & 0 & 0 & 7 & 7 & 0 & 19 & 37 & 61 & 0 \\
1131: \hline
1132: \beta_i^3 & 1 & 8 & 8 & 8 & 27 & 27 & 64 & 125 & (i-6)^3 \\
1133: \hline
1134: \end{array}
1135: \end{displaymath}
1136: Here we set $\alpha_i=0$ for $12 \le i \le 16$.
1137: So, their weights are defined as $1/\beta_i^3$.
1138:
1139: %% Given an item $p$ with size $x$,
1140: %% an efficient function $E_i(p)$ is defined as $W_i(p)/x^3$,
1141: %% then we have the following table.
1142: We first give two tables and then use them to calculate $\mathcal{P}(W)$.
1143: \begin{displaymath}
1144: \begin{array}{|c|l|l|l|l|l|l|l|}
1145: \hline
1146: i & (t_{i+1},t_i] & W_{1,1}(p) & E_{1,1}(p) & W_{2,1}(p) & E_{2,1}(p) &W_{2,2}(p) & E_{2,2}(p) \\
1147: \hline
1148: 1 &(0.7,1] & 1 & 2.9155 & 1 & 2.9155 & 1 & 2.9155 \\
1149: 2 &(0.65,0.7] & 1 & 3.65 & 1 & 3.65 & 1 & 3.65 \\
1150: 3 &(0.6,0.65] & 1 & 4.63 & 1 & 4.63 & 1 & 4.63 \\
1151: 4 &(0.5,0.6] & 1 & 8 & 0 & 0 & 1 & 8 \\
1152: 5 &(0.4,0.5] & 1/8 & 1.9532 & 1/8 & 1.9532 & 1/8 & 1.9532 \\
1153: 6 & (0.35,0.4]& 0.11 & 2.5656 & 0.1272 & 2.966 & 0.1272 & 2.966 \\
1154: 7 &(1/3,0.35] & 0.1 & 2.7 & 0.1286 & 3.472 & 0.1 & 2.7 \\
1155: 8 &(0.3,1/3] & 1/27 & 1.372 & 1/27 & 1.372 & 1/27 & 1.372 \\
1156: 9 &(1/4,0.3] & 0.025 & 1.6 &0.04211 & 2.6948 & 0.025 & 1.6 \\
1157: 10 &(1/5,1/4] & 0.0124& 1.55 & 0.01802 & 2.252 & 0.0124 & 1.55 \\
1158: 11 &(1/6,1/5] & 0.0068& 1.4688 & 0.0093 & 2 & 0.0068 & 1.4688 \\
1159: %% 12 &(1/7,1/6]& 1/216 & 1.59 & 1/216 & 1.59 & 1/216 & 1.59 & 1/216 & 1.59 \\
1160: %% 13 &(1/8,1/7]& 1/7^3 & 1.5 & 1/7^3 & 1.5 & 1/7^3 & 1.5 & 1/7^3 & 1.5 \\
1161: %% 14 &(1/9,1/8]& 1/8^3 & 1.43 & 1/8^3 & 1.43 & 1/8^3 & 1.43 & 1/8^3 & 1.43 \\
1162: %% 15 &(0.1,1/9]& 1/9^3 & 1.38 & 1/9^3 & 1.38 & 1/9^3 & 1.38 & 1/9^3 & 1.38 \\
1163: %% 16 &(1/11,0.1]& 0.001 & 1.331& 0.001 & 1.331 & 0.001 & 1.331 & 0.001 & 1.331 \\
1164: 12..17 &(0,1/6]& 1.59x^3 & 1.59 & 1.59x^3 & 1.59 & 1.59x^3 & 1.59 \\
1165: \hline
1166: \end{array}
1167: \end{displaymath}
1168:
1169:
1170: \paragraph{ Case 1:} $ W_1(X) \le 2.6852$.
1171:
1172: \noindent If $m_1+m_2 + m_3 + m_4 = 0$, i.e., no type $1,2,3,4$ items in $X$,
1173: then $m_6 + m_7 \le 8$,
1174: \[
1175: W_1(X) \le 0.11m_6+0.1m_7 + 1.96(1-0.35^3m_6-m_7/27)
1176: \le 2.3.
1177: \]
1178: Else $m_1 = 1$. Then $m_i=0$, where $2 \le i \le 8$,
1179: \[
1180: W_1(X) \le 1 + 1.6(1-0.7^3 ) = 2.0512.
1181: \]
1182: Else $m_2 = 1$. Then no type $1,3,4,5,6$ items in $X$ and $m_7 \le 7$,
1183: \[
1184: W_1(X) \le 1 + 0.1\times 7 + 1.6(1-0.65^3 - 7/27)
1185: \le 2.546.
1186: \]
1187: Else $m_3 =1$. Then no type $1,2,4,5$ items in $X$ and $m_6 + m_7 \le 7$,
1188: \[
1189: W_1(X)\le 1+ 0.11 m_6 + 0.1m_7 + 1.6(1-0.6^3 -0.35^3m_6-m_7/27) \le 2.5646.
1190: \]
1191: Else $m_4 = 1$. Then $m_1 + m_2 + m_3 = 0$ and $m_5 + m_6 + m_7 \le 7$,
1192: \[
1193: \begin{array}{lll}
1194: W_1(X) & \le & 1+ m_5/8 + 0.11 m_6 + 0.1m_7 \\
1195: & & + 1.6(1-0.5^3 - 0.4^3m_5 -0.35^3m_6-m_7/27) \\
1196: & < & 2.6852.
1197: \end{array}
1198: \]
1199: The last inequality follows from $m_7=7$ and $m_5=m_6=0$.
1200:
1201:
1202: \paragraph{ Case 2:} $ W_2(X) \le 2.6646$.
1203:
1204: \noindent If $m_1 +m_2 + m_3 + m_4 = 0$, i.e.,
1205: no type $1,2,3,4$ items in $X$, then $m_6 + m_7 \le 8$,
1206: \[
1207: W_2(X) = W_{2,2} \le 0.1272m_6+0.1m_7 + 1.96(1-0.35^3m_6-m_7/27)
1208: \le 2.4.
1209: \]
1210: Else $m_1 + m_2 = 1$ . Then no type $1,4,5,6$ items in $X$,
1211: \[
1212: W_2(X) = W_{2,2}(X) = W_1(X) \le 2.546.
1213: \]
1214: Else $m_3 = 1$. Then no type $1,2,4,5$ items in $X$ and $m_6 + m_7 \le 7$,
1215: \[
1216: W_2(X)\le 1+ 0.1272 m_6 + 0.1m_7 + 1.6(1-0.6^3 -0.35^3m_6-m_7/27) \le 2.6646.
1217: \]
1218: Else $m_4 =1$. Then no type $1,2,3$ items in $X$ and $m_6 + m_7 \le 7$,
1219: \[
1220: W_2(X) \le W_{2,1}(X) \le 0 + 0.1272m_6 + 0.1286 m_7 +
1221: 2.6948 (1 -0.35^3m_6-m_7/27 - 1/8) < 2.5595.
1222: \]
1223: The last inequality holds for $m_6 = 0$ and $m_7 = 7$.
1224:
1225: \begin{displaymath}
1226: \begin{array}{|c|l|l|l|l|l|l|l|}
1227: \hline
1228: i & (t_{i+1},t_i] & W_{3,1}(p) & E_{3,1}(p) & W_{3,2}(p) & E_{3,2}(p) & W_{4,1}(p) & E_{4,1}(p) \\
1229: \hline
1230: 1 &(0.7,1] & 1 & 2.9155 & 1 & 2.9155 & 1 & 2.9155 \\
1231: 2 &(0.65,0.7]& 1 & 3.65 & 1 & 3.65 & 0 & 0 \\
1232: 3 &(0.6,0.65]& 0 & 0 & 1 & 4.63 & 0 & 0 \\
1233: 4 &(0.5,0.6] & 0 & 0 & 1 & 8 & 0 & 0 \\
1234: 5 &(0.4,0.5] & 1/8 & 1.9532 & 1/8 & 1.9532 & 1/8 & 1.9532 \\
1235: 6 &(0.35,0.4]& 0.1272 & 2.966 & 0.1272 & 2.966 & 0.1272 &2.966 \\
1236: 7 &(1/3,0.35]& 0.1286 & 3.472 & 0.1286 & 3.472 & 0.03 & 0.81 \\
1237: 8 &(0.3,1/3] & 1/27 & 1.372 & 1/27 & 1.372 & 1/27 & 1.372 \\
1238: 9 &(1/4,0.3] & 0.04211 & 2.6948 & 0.025 & 1.6 & 0.04211& 2.6948 \\
1239: 10 &(1/5,1/4] & 0.01802 & 2.252 & 0.0124 & 1.55 &0.01802 & 2.252 \\
1240: 11 &(1/6,1/5] & 0.0093 & 2 & 0.0068 & 1.4688 & 0.0093 & 2 \\
1241: %% 12 &(1/7,1/6]& 1/216 & 1.59 & 1/216 & 1.59 & 1/216 & 1.59 & 1/216 & 1.59 \\
1242: %% 13 &(1/8,1/7]& 1/7^3 & 1.5 & 1/7^3 & 1.5 & 1/7^3 & 1.5 & 1/7^3 & 1.5 \\
1243: %% 14 &(1/9,1/8]& 1/8^3 & 1.43 & 1/8^3 & 1.43 & 1/8^3 & 1.43 & 1/8^3 & 1.43 \\
1244: %% 15 &(0.1,1/9]& 1/9^3 & 1.38 & 1/9^3 & 1.38 & 1/9^3 & 1.38 & 1/9^3 & 1.38 \\
1245: %% 16 &(1/11,0.1]& 0.001 & 1.331& 0.001 & 1.331 & 0.001 & 1.331 & 0.001 & 1.331 \\
1246: 12..17 &(0,1/6]& 1.59x^3 & 1.59 & 1.59x^3 & 1.59 & 1.59x^3 & 1.59 \\
1247: \hline
1248: \end{array}
1249: \end{displaymath}
1250:
1251: \paragraph{ Case 3:} $ W_3(X) \le 2.646$.
1252:
1253: \noindent If $m_1+ m_2 + m_3 + m_4 = 0$, i.e.,
1254: no type $1,2,3,4$ items in $X$,
1255: then $m_6 + m_7 \le 8$,
1256: \[
1257: W_3(X) = W_{3,2} \le 0.1272m_6+0.1286m_7 + 1.96(1-0.35^3m_6-m_7/27)
1258: \le 2.41.
1259: \]
1260: Else $m_1 =1$. Then $m_i=0$, where $2 \le i \le 8$,
1261: \[
1262: W_3(X) = W_{3,2}(X) = W_1(X) \le 2.0512.
1263: \]
1264:
1265: Else $m_2 = 1$. Then no type $1,3,4,5,6$ items in $X$ and $m_7 \le 7$,
1266: \[
1267: W_3(X) = W_{3,2}(X) \le 1 + 0.1286\times 7 + 1.6(1-0.65^3 - 7/27)
1268: \le 2.646.
1269: \]
1270:
1271: Else $m_3 + m_4 = 1$.
1272: Then no type $1,2$ items in $X$.
1273: \[
1274: W_3(X) \le W_{3,1}(X) = W_{2,1}(X) \le 2.5595.
1275: \]
1276:
1277: \paragraph{ Case 4:} $ W_4(X) \le 2.63$.
1278:
1279: \noindent If $m_1+ m_2 + m_3 + m_4 = 0$, i.e.,
1280: no type $1,2,3,4$ items in $X$, $m_6 + m_9 \le 27$ and $m_6 \le 8$,
1281: \[
1282: W_4(X) \le 0.1272m_6+0.04211m_9 + 2.252(1-0.35^3m_6-m_9/64)
1283: \le 2.63.
1284: \]
1285: %% The last inequality holds for $m_6 = 8$ and $m_9 = 19$.
1286:
1287: Else $m_1 =1$. Then $m_i=0$, where $2 \le i \le 8$.
1288: And $m_9 \le 19$,
1289: \[
1290: W_4(X) \le 1 + 19 \times 0.04211 + 2.252(1-0.7^3 -19/64)
1291: \le 2.62.
1292: \]
1293: Else $m_2 + m_3 + m_4 = 1$.
1294: Then $m_6 \le 7$,
1295: \[
1296: W_4(X) \le 0 + 0.1272 m_6 +
1297: 2.6948 (1 -0.35^3m_6 - 1/8) < 2.4396.
1298: \]
1299:
1300:
1301: %% \noindent {\bf Case 1}:
1302: %% There are no type $1,2,3,4$ items in $X$,
1303: %% i.e., $m_i=0$ for $1 \le i \le 4$.
1304: %% Since $m_6 + m_9 \le 27$ and $m_6 \le 8$,
1305: %% \[
1306: %% W_4(X) \le 0.1272m_6+0.04211m_9 + 2.252(1-0.35^3m_6-m_9/64)
1307: %% \le 2.63.
1308: %% \]
1309: %% The last inequality holds for $m_6 = 8$ and $m_9 = 19$.
1310:
1311: %% On the other hand, $W_1(X)\le W_2(X) \le W_3(X)$ and $m_6 + m_7 \le 8$,
1312: %% \[
1313: %% W_3(X) \le 0.1272m_6+0.1286m_7 + 1.96(1-0.35^3m_6-m_7/27)
1314: %% \le 2.41.
1315: %% \]
1316: %% The last inequality holds for $m_6 = 0$ and $m_7 = 8$.
1317: %% So, we have $\mathcal{P}(W) \le \max\{W_i(X)\} \le 2.63$.
1318:
1319: %% \noindent {\bf Case 2}: $m_1 =1$,
1320: %% there is one type $1$ item in $X$. So, $m_i=0$, where $2 \le i \le 8$.
1321: %% And $m_9 \le 19$, then
1322: %% \[
1323: %% \mathcal{P}(W) \le 1 + 19 \times 0.04211 + 2.252(1-0.7^3 -19/64)
1324: %% \le 2.62.
1325: %% \]
1326:
1327: %% \noindent {\bf Case 3}: $m_2 =1$, one type $2$ item in $X$.
1328: %% %% So, $m_i=0$ and $m_1=0$, where $3 \le i \le 6$.
1329: %% So no type $1,3,4,5,6$ items in $X$.
1330: %% Then as proved in Case 1, $ W_4(X) \le 2.63$.
1331: %% Since no type $5,6$ items in $X$ and $m_7 \le 7$,
1332: %% then
1333: %% \[
1334: %% \max\{W_1(X),W_2(X), W_3(X)\} \le 1 + 0.1286\times7 + 1.6(1-0.65^3 - 7/27)
1335: %% < 2.646.
1336: %% \]
1337: %% So, $\mathcal{P}(W) < 2.646$.
1338:
1339: %% \noindent {\bf Case 4}: $m_3 =1$, one type $3$ item in $X$.
1340: %% So no type $1,2,4,5$ items in $X$.
1341: %% Then by the proof in Case 1
1342: %% \[
1343: %% \max\{W_3(X),W_4(X)\} < 2.63.
1344: %% \]
1345: %% Else we just need to consider $W_2(X)$, for $W_1(X) \le W_2(X)$.
1346: %% Since $m_1 + m_2 + m_4 +m_5 =0$ and $m_6 + m_7 \le 7$,
1347: %% \[
1348: %% W_2(X)\le 1+ 0.1272 m_6 + 0.1m_7 + 1.6(1-0.6^3 -0.35^3m_6-m_7/27) \le 2.6646.
1349: %% \]
1350: %% The last inequality follows from $m_6=7$ and $m_7=0$.
1351: %% So, $\mathcal{P}(W) < 2.6646$.
1352:
1353: %% \noindent {\bf Case 5}: $m_4 =1$, one type $4$ item in $X$.
1354: %% So no type $1,2,3$ items in $X$.
1355: %% Then by the proof in Case 1
1356: %% \[
1357: %% \max\{W_2(X), W_3(X),W_4(X)\} < 2.63.
1358: %% \]
1359:
1360: %% Else $m_5 + m_6 + m_7 \le 7$ and $m_1 + m_2 + m_3 = 0$,
1361: %% \[
1362: %% \begin{array}{lll}
1363: %% W_1(X) & \le & 1+ m_5/8 + 0.11 m_6 + 0.1m_7 \\
1364: %% & & + 1.6(1-0.5^3 - 0.4^3m_5 -0.35^3m_6-m_7/27) \\
1365: %% & < & 2.6852.
1366: %% \end{array}
1367: %% \]
1368: %% The last inequality follows from $m_7=7$ and $m_5=m_6=0$.
1369: So, $\mathcal{P}(W) < 2.6852$.
1370: \end{proof}
1371: %%\subsection{Online cube packing}
1372:
1373: %% \begin{theorem}
1374: %% When $d$ is a fixed integer,
1375: %% the asymptotic performance ratio of $\mathcal{A}$
1376: %% for $d$-dimensional hypercube packing is O(d).
1377: %% \end{theorem}
1378: %% \begin{proof}
1379: %% In algorithm $\mathcal{A}$,
1380: %% we set $M = 6$, i.e., an item $R$ is {\em small}
1381: %% if its size is at most $1/d$.
1382: %% And we also set $t_i = 1/i$ and $\alpha_i =0$ for $ 1 \le i \le d$.
1383: %% Then we define a weighting fuction $W(p)$ for an item $p$ with size $x$.
1384: %% we define weighting function $W_i$ for case $i$ for $1 \le i \le 5$.
1385: %% \begin{displaymath}
1386: %% \begin{array}{ll}
1387: %% W(p) = & \left\{
1388: %% %% { \setlength \arrayrowsep{1mm}
1389: %% \begin{array}{ll}
1390: %% \frac{1}{i^d} & \textrm{if } 1/(i+1) \le x \le 1/i, \textrm{ for } 1 \le i \le d-1. \\
1391: %% \vspace{4pt}
1392: %% \frac{x^d (d+1)^d}{d^2 -1} & \textrm{if } x \le 1/d.
1393: %% \end{array}
1394: %% \right.
1395: %% \end{array}
1396: %% \end{displaymath}
1397: %% We see Lemma \ref{lemma:upper} still holds, i.e.,
1398: %% to get an upper bound for algorithm $\mathcal{A}$'s asymptotic worst case
1399: %% ratio, we just need to estimate the upper bound of
1400: %% \[
1401: %% W(X) = \sum_{p \in X} W(p),
1402: %% \]
1403: %% where $X$ is a feasible set in which all items can be placed
1404: %% in a d-dimensional bin.
1405: %% From the above definitions for $t_i$,
1406: %% we get $d-1$ types of large items.
1407: %% Define $m_i$ as the number of type $i$ items in $X$,
1408: %% where $ 1 \le i _\le d -1$.
1409: %% So, we have
1410: %% \begin{equation}
1411: %% \label{eqn:mi}
1412: %% m_i \le i^d,
1413: %% \end{equation}
1414: %% since every type $i$ item has size in $(1/(i+1),1/i]$.
1415: %% Let $V_s$ and $V_l$ be the total volume of large items and small items,
1416: %% respectively.
1417: %% Let $X = X_l \cup X_s$,
1418: %% where $X_l$ and $X_s$ are the subsets of $X$ for large items and small
1419: %% items, respectively.
1420: %% Then by (\ref{eqn:mi})
1421: %% \[
1422: %% W(X_l) = \sum_{i=1}^{d-1} \frac{m_i}{i^d} \le d-1.
1423: %% \]
1424: %% On the other hand,
1425: %% \[
1426: %% W(X_s) = \sum_{p \in X_s} W(p) = \sum_{p \in X_s}\frac{x^d(d+1)^d}{d^d-1}
1427: %% = V_s\frac{(d+1)^d}{d^d-1} < d+2.
1428: %% \]
1429: %% The last inequality holds for $V_s \le 1$ and
1430: %% \[
1431: %% \frac{(d+1)^d}{d^d-1} = \frac{\sum_{i=0}^{d}\binom{d}{i}d^{d-i}}{d^d-1}
1432: %% < \frac{(d+1)d^d}{d^d-1} < d+2.
1433: %% \]
1434: %% Hence $W(X)= W(X_l) + W(X_s) = O(d)$.
1435: %% \end{proof}
1436: \section{Concluding Remarks}
1437: In this page, we reduce the gaps between the upper and lower bounds of
1438: online square packing and cube packing.
1439: But the gaps are still large.
1440: It seems possible to use computer proof as the one in \cite{s02}
1441: to get a more precise upper bound.
1442: But, the analysis becomes more complicated and more difficult than
1443: the one in \cite{s02},
1444: since we are faced to solve a two dimensional knapsack problem, rather
1445: than one dimensional knapsack problem \cite{s02}.
1446: So, how to reduce the gaps is a challenging open problem.
1447:
1448: \begin{thebibliography}{99}
1449: \baselineskip 11.5pt
1450: \bibitem{B79}
1451: D.J. Brown,
1452: A lower bound for on-line one-dimensional bin packing algorithms,
1453: {\em Techincal report R864}, Coordinated Sci. Lab., Urbana, Illinois, 1979.
1454:
1455:
1456: \bibitem{CGJ97}
1457: E.G. Coffman, M.R. Garey and D.S. Johnson,
1458: Approximation algorithms for bin packing: a survey.
1459: In Approximation Algorithms for NP-hard Problems,
1460: D. Hochbaum, Ed. PWS, Boston, MA, 1997, chapter 2.
1461:
1462: \bibitem{CR89}
1463: D. Coppersmith, P. Paghavan,
1464: Multidimensional on-line bin packing:
1465: Algorithms and worst case analysis,
1466: {\em Oper. Res. Lett.} 8:17-20, 1989.
1467:
1468:
1469: \bibitem{CV93}
1470: J. Csirik, A. van Vliet,
1471: An on-line algorithm for multidimensional bin packing,
1472: {\em Oper. Res. Lett.} 13: 149-158, 1993.
1473:
1474: \bibitem{cw97}
1475: J. Csirik and G.J. Woeginger, Shelf algorithm for on-line strip
1476: packing, {\em Information Processing Letters} 63, 171-175, 1997.
1477:
1478: \bibitem{ES04}
1479: L. Epstein, R. van Stee,
1480: Optimal online bounded space multidimensional packing,
1481: {\em SODA 2004}: 214-223.
1482:
1483: \bibitem{ES05a}
1484: L.Epstein, R. van Stee,
1485: Optimal Online Algorithms for Multidimensional Packing Problems.
1486: {\em SIAM J. Computing}, 35(2): 431-448, 2005.
1487:
1488: \bibitem{ES05b}
1489: L.Epstein, R. van Stee,
1490: Online square and cube packing,
1491: { \em Acta Inf.} 41(9): 595-606, 2005.
1492:
1493:
1494: \bibitem{JDUGG74}
1495: D.S. Johnson, A.J. Demers, J.D. Ullman, M. R. Garey, R. L. Graham,
1496: Worst-Case performance bounds for simple one-dimensional packing algorithms,
1497: {\em SIAM J. Comput.} 3(4): 299-325, 1974.
1498:
1499: \bibitem{L80}
1500: F.M. Liang,
1501: A lower bound for online bin packing,
1502: {\em Information processing letters} 10,76-79,1980.
1503:
1504: \bibitem{LL85}
1505: C.C. Lee and D.T. Lee, A simple on-line bin-packing algorihtm,
1506: {\em J. ACM} 32, 562-572, 1985.
1507:
1508: \bibitem{MW03}
1509: F.K. Miyazawa, Y. Wakabayashi,
1510: Cube packing,
1511: {\em Theor. Comput. Sci.} 1-3(297): 355-366, 2003.
1512:
1513: \bibitem{RBLL89}
1514: P.V. Ramanan, D.J. Brown, C.C. Lee, and D. T. Lee, On-line bin
1515: packing in linear Time, {\em J. Algorithms} 10, 305-326, 1989.
1516:
1517: \bibitem{s02}
1518: S.S. Seiden, On the online bin packing problem, {\em J. ACM} 49,
1519: 640-671, 2002.
1520:
1521: \bibitem{SS03}
1522: S.S. Seiden, R. van Stee,
1523: New bounds for multidimensional packing,
1524: {\em Algorithmica} 36(3): 261-293, 2003.
1525:
1526: \bibitem{Vliet92}
1527: A. van Vliet, An improved lower bound for on-line bin packing
1528: algorithms, {\em Inform. Process. Lett.} 43, 277-284,1992.
1529:
1530: \bibitem{Yao80} A.C.-C. Yao, New Algorithms for Bin Packing,
1531: {\em J. ACM} 27, 207-227, 1980.
1532:
1533: %% \bibitem{B85}
1534: %% B.S. Baker, A new proof for the first-fit decreasing bin-packing
1535: %% algorithm. {\em J. Algorithms} 6, 49-70, 1985.
1536:
1537: %% \bibitem{BBK81}
1538: %% B.S. Baker, D.J. Brown, and H.P. Katseff, A 5/4 algorithm for
1539: %% two-dimensional packing. {\em J. Algorithms} 2, 348-368, 1981.
1540:
1541: %% \bibitem{BS83}
1542: %% B.S. Baker and J.S. Schwarz, Shelf algorithms for two-dimensional
1543: %% packing problems, {\em SIAM J. Comput.} 12, 508-525, 1983.
1544:
1545: %% \bibitem{CGJT80}
1546: %% E.G. Coffman, M.R. Garey, D.S. Johnson, and R.E. Tarjan,
1547: %% Performance bounds for level oriented two dimensional packing
1548: %% algorithms, {\em SIAM J. Comput.} 9, 808-826, 1980.
1549:
1550: %% \bibitem{cw97}
1551: %% J. Csirik and G.J. Woeginger, Shelf algorithm for on-line strip
1552: %% packing, {\em Information Processing Letters} 63, 171-175, 1997.
1553:
1554: %% \bibitem{Golan81}
1555: %% I. Golan, Performance bounds for orthogonal, oriented
1556: %% two-dimensional packing algorithms, {\em SIAM J. Comput.} 10,
1557: %% 571-582, 1981.
1558:
1559: %% \bibitem{J73}
1560: %% D.S. Johnson, Near-optimal bin-packing algorithms, {\em doctoral
1561: %% thesis, M.I.T., Cambridge, Mass.}, 1973.
1562:
1563: %% \bibitem{kk82}
1564: %% N. Karmarkar and R.M. Karp, An efficient approximation scheme for
1565: %% the one-dimensional bin-packing problem, In {\em Proc. 23rd Annual
1566: %% IEEE Symp. Found. Comput. Sci.}, 312-320, 1982.
1567:
1568: %% \bibitem{KR00}
1569: %% C. Kenyon and E.R\'emila, A near-optimal solution to a
1570: %% two-dimensional cutting stock problem, {\em Mathematics of
1571: %% Operations Research} 25, 645-656, 2000.
1572:
1573:
1574:
1575:
1576:
1577: %% \bibitem{R91} M.B. Richey, Improved bounds for harmonic-based bin packing
1578: %% algorithms, {\em Discrete Appl. Math.} 34, 203-227, 1991.
1579:
1580:
1581:
1582:
1583: %% \bibitem{SW05}
1584: %% S.S. Seiden and G.J. Woeginger, The two-dimensional cutting stock
1585: %% problem revisited, {\em Math. Program.} Ser.A 102, 519-530, 2005.
1586:
1587: %% \bibitem{ullman71}
1588: %% J.D. Ullman, The performance of a memory allocation algorithm. {\em
1589: %% Tech. Rep. 100, Princeton University, Princeton, N.J.,Oct.}, 1971.
1590:
1591:
1592:
1593:
1594:
1595: %% \bibitem{Y91}
1596: %% M. Yue, A simple proof of the inequality FFD(L) $\le$ 11/9OPT(L)
1597: %% +1, $\forall L$ for the FFD bin-packing algorithm, {\em Acta
1598: %% mathematicae applicatae sinica} 7, 321-331, 1991.
1599:
1600: \end{thebibliography}
1601:
1602:
1603: %% \normalsize
1604: %% \newpage
1605: %% \begin{center}
1606: %% {\bf Appendix}
1607: %% \end{center}
1608:
1609: %% \begin{table} [h]
1610:
1611: %% \caption{ $W_{1,1}(p)$, $W_{2,1}(p)$, $W_{2,2}(p)$}
1612: %% \label{table:t1}
1613: %% \end{table}
1614:
1615: %% \begin{table} [ht]
1616:
1617: %% \caption{ $W_{3,1}(p)$, $W_{3,2}(p)$ and $W_{4,1}(p)$}
1618: %% \label{table:t2}
1619: %% \end{table}
1620:
1621: \end{document}
1622: