Online scheduling of precedence-constrained jobs on a single machine
5
0
0
全文
(2). .ohei +atanoE
(3). 6huji .ijimaF
(4). (iji 7akimotoG
(5). Abstract: :H FRQVLGHU DQ RQOLQH MRE VFKHGXOLQJ SUREOHP RI SUHFHGHQFHFRQVWUDLQHG MREV RQ D VLQJOH PDFKLQH ,Q WKLV. SUREOHP WKH SOD\HU LV VXSSRVHG WR GHWHUPLQH D VFKHGXOH RI n ¿[HG MREV DW HDFK WULDO XQGHU WKH FRQVWUDLQWV WKDW VRPH MREV KDYH KLJKHU SULRULW\ WKDQ RWKHU MREV LQ HDFK SHUPXWDWLRQ 7KH JRDO LV WR PLQLPL]H WKH VXP RI FRPSOHWLRQ WLPHV RYHU DOO MREV DQG T GD\V XQGHU SUHFHGHQFH FRQVWUDLQWV :H SURSRVH DQ RQOLQH MRE VFKHGXOLQJ DOJRULWKP ZKLFK SUHGLFWV DOPRVW DV ZHOO DV WKH EHVW NQRZQ RfflLQH DSSUR[LPDWLRQ DOJRULWKPV LQ KLQGVLJKW )XUWKHUPRUH RXU DOJRULWKP UXQV LQ O n
(6) WLPH IRU HDFK WULDO. ,QWURGXFWLRQ :H FRQVLGHU DQ RQOLQH YHUVLRQ RI MRE VFKHGXOLQJ ZLWK D VLQJOH PDFKLQH XQGHU VRPH SUHFHGHQFH FRQVWUDLQWV $VVXPH WKDW WKHUH LV D VLQJOH SURFHVVRU DQG n ¿[HG MREV (YHU\ GD\ t ZH GHWHUPLQH D VFKHGXOH KDYH WR EH FRQVLVWHQW ZLWK SUHFHGHQFH FRQVWUDLQWV 7KHQ DIWHU SURFHVVLQJ DOO n MREV DFFRUGLQJ WR WKH VFKHGXOH WKH SURFHVV LQJ WLPH RI HDFK MRE LV UHYHDOHG 7KH JRDO LV WR PLQLPL]H the sum of the completion time RYHU DOO MREV DQG T GD\V XQGHU WKH ¿[HG SUHFHGHQFH FRQVWUDLQWV ZKHUH WKH FRPSOHWLRQ WLPH RI MRE i DW GD\ t LV WKH VXP RI SURFHVVLQJ WLPHV RI MREV SULRU WR i DQG WKH SURFHVV LQJ WLPH RI MRE i ,Q WKLV SDSHU ZH UHSUHVHQW D VFKHGXOH RI MREV E\ D SHUPX WDWLRQ )RU H[DPSOH DW GD\ t ZH SURFHVV MREV DFFRUGLQJ WR D SHUPXWDWLRQ , , ,
(7) DQG HDFK SURFHVVLQJ WLPH LV JLYHQ DV t = t, , t, , t, , t,
(8) 1RWH WKDW WKH FRPSRQHQW RI SHUPXWD WLRQ , , ,
(9) UHSUHVHQWV WKH SULRULW\ RI HDFK MRE 7KDW LV MREV ZLWK KLJKHU SULRULW\ DUH SURFHVVHG HDUOLHU 7KHUHIRUH MREV , , DQG DUH SURFHVVHG VHTXHQWLDOO\ 7KH FRPSOHWLRQ WLPH RI MREV i = , , , DUH t, t, +t, t, +t, +t, DQG t, +t, +t, +t, UHVSHFWLYHO\ 6R DQ LQQHU SURGXFW , , ,
(10) · t H[DFWO\ FRUUH VSRQGV WR WKH VXP RI WKH FRPSOHWLRQ WLPH DW GD\ t $ SHUPXWDWLRQ σ RYHU WKH VHW >n@ = {, . . . , n} RI n ¿[HG REMHFWV LV D ELMHFWLYH IXQFWLRQ IURP >n@ WR >n@ $QRWKHU UHSUHVHQWDWLRQ RI D SHUPXWDWLRQ σ RYHU WKH VHW >n@ LV WR GHVFULEH LW DV DQ nGLPHQVLRQDO YHFWRU LQ >n@n GH¿QHG DV σ = σ
(11) , . . . , σ n
(12)
(13) (J , , ,
(14) LV D UHSUHVHQWDWLRQ RI D SHUPXWDWLRQ IRU n = /HW S n EH WKH VHW RI DOO SHUPXWDWLRQV RYHU >n@ LH S n = {σ ∈ >n@n | σ LV D SHUPXWD WLRQ RYHU >n@} ,Q SDUWLFXODU WKH FRQYH[ KXOO RI DOO SHUPXWDWLRQV LV FDOOHG SHUPXWDKHGURQ GHQRWHG DV Pn :H DVVXPH D VHW RI SUHFHGHQFH FRQVWUDLQWV LQ SHUPXWDWLRQV D
(15) E
(16) F
(17) G
(18). 'HSDUWPHQW RI ,QIRUPDWLFV .\XVKX 8QLYHUVLW\ WDNDKLURIXMLWD#LQIN\XVKXXDFMS KDWDQR#LQIN\XVKXXDFMS NLMLPD#LQIN\XVKXXDFMS HLML#LQIN\XVKXXDFMS. F ,QIRUPDWLRQ 3URFHVVLQJ 6RFLHW\ RI -DSDQ . 7KH VHW A RI SUHFHGHQFH FRQVWUDLQWV LV JLYHQ DV A = { ik , jk
(19) ∈ >n@ × >n@ | ik jk , k = , . . . , m} PHDQLQJ WKDW REMHFW ik LV SUH IHUUHG WR REMHFW jk 7KH VHW A LQGXFHV WKH VHW GH¿QHG E\ OLQHDU FRQVWUDLQWV 3UHFRQV A
(20) = {p ∈ Rn+ | pi ≥ p j IRU i, j
(21) ∈ A} :H IXUWKHU DVVXPH WKDW WKHUH H[LVWV D OLQHDU RUGHULQJ FRQVLVWHQW ZLWK A ,Q RWKHU ZRUGV ZH DVVXPH WKHUH H[LVWV D SHUPXWDWLRQ σ ∈ S n ∩ 3UHFRQV A
(22) ,Q WKLV SDSHU ZH FRQVLGHU WKH IROORZLQJ RQOLQH VFKHGXOLQJ SUREOHP RYHU S n ∩ 3UHFRQV A
(23) )RU HDFK WULDO t = , . . . , T L
(24) WKH SOD\HU SUHGLFWV D SHUPXWDWLRQ σt ∈ S n ∩ 3UHFRQV A
(25) LL
(26) WKH DGYHUVDU\ UHWXUQV D ORVV YHFWRU t ∈ >, @n DQG LLL
(27) WKH SOD\HU LQ FXUV ORVV σt · t 7KH JRDO RI WKH SOD\HU LV WR PLQLPL]H WKH αUHJUHW IRU VRPH VPDOO α ≥ α5HJUHW =. T t=. σt · t − α. PLQ. σ∈S n ∩3UHFRQV A
(28). T . σ · t .. t=. ,Q WKLV SDSHU ZH SURSRVH DQ RQOLQH VFKHGXOLQJ DOJRULWKP RYHU √ Pn ∩3UHFRQV A
(29) ZKRVH αUHJUHW LV O n T
(30) IRU α = −/ n+
(31) )RU HDFK WULDO RXU DOJRULWKP UXQV LQ O n
(32) WLPH )XUWKHU ZH √ VKRZ WKDW WKH ORZHU ERXQG RI WKH UHJUHW LV Ω n T
(33) ,Q DG GLWLRQ ZH SURYH WKDW WKHUH LV QR SRO\QRPLDO WLPH DOJRULWKP ZLWK √ αUHJUHW poly n, m
(34) T ZLWK α < − / n +
(35) XQOHVV WKHUH H[ LVWV D UDQGRPL]HG DSSUR[LPDWLRQ DOJRULWKP ZLWK DSSUR[LPDWLRQ α < − / n +
(36) IRU WKH FRUUHVSRQGLQJ RfflLQH SUREOHP ZKLFK ZH GLVFXVV ODWHU
(37) 6R IDU WKH VWDWHRIWKHDUW DSSUR[LPDWLRQ DOJR ULWKPV KDYH DSSUR[LPDWLRQ UDWLR − / n +
(38) DQG LW LV DQ RSHQ SUREOHP WR ¿QG DQ DSSUR[LPDWLRQ DOJRULWKP ZLWK EHWWHU UDWLR >@ 7KHUHIRUH RXU DOJRULWKP LV RSWLPDO DPRQJ DQ\ SRO\QRPLDO DOJR ULWKPV XQOHVV WKH RSHQ SUREOHP LV SRVLWLYHO\ VROYHG 7KH FRUUHVSRQGLQJ RfflLQH SUREOHP KDV EHHQ H[WHQVLYHO\ LQ YHVWLJDWHG LQ WKH OLWHUDWXUH 7KH SUREOHP LV JLYHQ D ORVV YHFWRU ∈ >, @n DQG WKH VHW RI SUHFHGHQFH FRQVWUDLQWV A DV LQSXWV WR RXWSXW D SHUPXWDWLRQ σ ∈ S n ∩ 3UHFRQV A
(39) ZKLFK PLQLPL]HV WKH LQQHU SURGXFW σ · LH WKH VXP RI FRPSOHWLRQ WLPHV 0RUH JHQ HUDOO\ WKH SUREOHP RI PLQLPL]LQJ WKH ZHLJKWHG VXP RI FRPSOH 1.
(40) Vol.2014-AL-150 No.9 2014/11/20. ,36- 6,* 7HFKQLFDO 5HSRUW. WLRQ WLPHV DUH W\SLFDOO\ FRQVLGHUHG ,W LV NQRZQ WKDW WKH SUREOHP LV 13KDUG >@ >@ 6HYHUDO − O /n
(41) DSSUR[LPDWLRQ DOJR ULWKPV DUH SURSRVHG 6FKXO] > @ +DOO HW DO >@ &KXGDN DQG +RFKEDXP >@ 0DUJRW HW DO >@ DQG &KHNXUL DQG 0RWZDQL >@
(42) )RU IXUWKHU GHYHORSPHQWV VHH HJ >@ >@ 7KHUH DUH UHODWHG UHVHDUFKHV RQ RQOLQH VFKHGXOLQJ RYHU WKH SHU PXWDKHGURQ 7KH ¿UVW UHVXOW ZLWKRXW SUHFHGHQFH FRQVWUDLQWV LV SURSRVHG E\ <DVXWDNH HW DO >@ $LORQ SURSRVHG DQRWKHU RQOLQH RSWLPL]DWLRQ DOJRULWKP ZLWK DQ LPSURYHG UHJUHW ERXQG>@ 6XH KLUR HW DO >@ H[WHQGHG <DVXWDNH HW DO¶V UHVXOW WR WKH VXEPRGXODU EDVH SRO\KHGURQ ZKLFK FDQ EH XVHG IRU QRW RQO\ SHUPXWDWLRQV EXW DOVR RWKHU FRPELQDWRULDO REMHFWV VXFK DV VSDQQLQJ WUHHV ,W LV SRVVLEOH WR REWDLQ RQOLQH VFKHGXOLQJ DOJRULWKPV XVLQJ ³RfflLQHWRRQOLQH´ FRQYHUVLRQ WHFKQLTXHV %\ XVLQJ FRQYHUVLRQ PHWKRG RI .DNDGH HW DO >@ RU )XMLWD HW DO >@ ZH FDQ FRQVWUXFW RQOLQH RSWLPL]DWLRQ DOJRULWKPV ZLWK αUHJUHW FORVH WR RXUV %XW ZLWK WKH PHWKRG RI .DNDGH HW DO >@ WKH UHVXOWLQJ DOJRULWKP WDNHV WLPH OLQHDU LQ T ZKLFK LV QRW GHVLUDEOH :LWK WKH PHWKRG RI )XMLWD HW DO >@ WKH UXQQQLQJ WLPH SHU WULDO LV poly n, /ε
(43) ZKLK LV LQGHSHQGHQW RI T EXW GHSHQGV RQ /ε DQG LWV αUHJUHW LV SURYHG IRU α = − / n +
(44) + ε ZKLFK LV LQIHULRU WR RXUV. 3UHOLPLQDULHV )RU DQ\ ¿[HG SRVLWLYH LQWHJHU n OHW >n@ E\ WKH VHW {, . . . , n} 7KH permutahedron Pn LV WKH FRQYH[ KXOO RI WKH VHW RI SHUPXWD WLRQV S n ,W LV NQRZQ WKDW Pn FDQ EH UHSUHVHQWHG DV WKH VHW RI SRLQWV p ∈ Rn+ VDWLVI\LQJ i∈S pi ≤ |Si=| n + − i
(45) IRU DQ\ S ⊂ >n@ DQG n i= pi = n n +
(46) / )RU UHIHUHQFHV RI WKH SHUPXWDKHGURQ VHH HJ >@ >@ :H ZLOO XVH D JHRPHWULF SURSHUW\ RI %UHJPDQ GLYHUJHQFH ZKLFK LV NQRZQ DV *HQHUDOL]HG 3\WKDJRUHDQ 7KHRUHP :H VKRZ D YHUVLRQ RI WKH WKHRUHP DGDSWHG IRU (XFOLGHDQ QRUP 7KHRUHP %UHJPDQ>@ >@
(47) Let C ⊂ Rn+ be any convex set. Let q be any point in Rn+ and let p = LQI p ∈C ||p − q|| . Then, it holds for any r ∈ C that. 2*'
(48) >@ LQ RQOLQH OHDUQLQJ OLWHUDWXUH 2*' FRQVLVWV RI WKH DGGLWLYH XSGDWH RI ZHLJKW YHFWRUV DQG WKH SURMHFWLRQ WR VRPH FRQ YH[ VHW RI LQWHUHVW ,Q RXU FDVH WKH FRQYH[ VHW LV Pn ∩ 3UHFRQV A
(49) 8VLQJ WKHVH SURFHGXUHV WKH UHJUHW ERXQG RI 2*' FDQ EH SURYHG √ WR EH O n T
(50) 6R DSSDUHQWO\ RXU VXFFHVVLYH SURMHFWLRQV VHHP UHGXQGDQW DQG RQO\ RQH SURMHFWLRQ WR Pn ∩ 3UHFRQV A
(51) ZRXOG VXI ¿FH 7KH SUREOHP RI WKH VWDQGDUG DSSURDFK LV WKDW WKH SURMHFWLRQ RQWR Pn ∩ 3UHFRQV A
(52) ORRNV QRW WUDFWDEOH VLQFH LW GHDOV H[SRQHQ WLDOO\ PDQ\ OLQHDU FRQVWUDLQWV /DWHU ZH ZLOO VKRZ WKDW WKH VXF FHVVLYH SURMHFWLRQV DUH WKH NH\V WR DQ HffiFLHQW LPSOHPHQWDWLRQ RI RXU DOJRULWKP )LUVW ZH SURYH DQ αUHJUHW ERXQG RI WKH SURSRVHG DOJRULWKP DQG WKHQ ZH VKRZ WKDW RXU DOJRULWKP FDQ EH HffiFLHQWO\ UHDOL]HG $OJRULWKP 3HUP/HDUQ3UHF
(53) /HW p = n +
(54) /, . . . , n +
(55) /
(56) ∈ >, n@n
(57) )RU t = , . . . , T D
(58) 5XQ 5RXQGLQJ pt
(59) DQG JHW σt ∈ S n VXFK WKDW σt ≤ − / n +
(60)
(61) pt E
(62) ,QFXU D ORVV σt · t F
(63) 8SGDWH pt+ DV pt+ = pt − ηt G
(64) /HW pt+ EH WKH (XFOLGHDQ SURMHFWLRQ RQWR WKH VHW 3UHFRQV A
(65) LH pt+ = DUJ PLQp∈3UHFRQV A
(66) ||p − pt+ || H
(67) /HW pt+ EH WKH SURMHFWLRQ RI pt+ RQWR WKH VHW Pn ∩ 3UHFRQV A
(68) WKDW LV pt+ = DUJ LQI p∈Pn ∩3UHFRQV A
(69) ||p − pt+ || . :H VWDUW WKH DQDO\VLV RI 3HUP/HDUQ3UHF ZLWK WKH IROORZLQJ OHPPD 7KH OHPPD JXDUDQWHHV WKH ³SURJUHVV ³ RI pt WRZDUGV DQ\ YHFWRU LQ Pn ∩3UHFRQV A
(70) ZKLFK LV PHDVXUHG E\ (XFOLGHDQ QRUP VTXDUHG /HPPD For any q ∈ Pn ∩ 3UHFRQV A
(71) and for any t ≥ , ||q − pt || − ||q − pt+ || ≥ η q − pt
(72) · t − η ||t || . Proof. %\ XVLQJ *HQHUDOL]HG 3\WKDJRUHDQ 7KHRUHP LQ 7KHR UHP ||q − pt+ || ≥ ||q − pt+ || + ||pt+ − pt+ ||. ||r − q|| ≥ ||r − p|| + ||p − q|| . Further, this inequality becomes an equality if C is an affine set.. $OJRULWKP ,Q WKLV VHFWLRQ ZH SURSRVH RXU DOJRULWKP 3HUP/HDUQ3UHF DQG SURYH LWV αUHJUHW ERXQG 0DLQ 6WUXFWXUH 7KH GHVFULSWLRQ RI 3HUP/HDUQ3UHF LV VKRZQ LQ $OJRULWKP 7KH DOJRULWKP PDLQWDLQV D ZHLJKW YHFWRU pt LQ Rn+ ZKLFK UHS UHVHQWV D ³PL[WXUH´ RI SHUPXWDWLRQV LQ S n $W HDFK WULDO t LW ³URXQGV´ D YHFWRU pt LQWR D SHUPXWDWLRQ σt VR WKDW σt ≤ αpt IRU VRPH α > 7KLV SURFHGXUH LV GRQH E\ 5RXQGLQJ ZKLFK ZH ZLOO VKRZ WKH GHWDLOV LQ WKH QH[W VHFWLRQ $IWHU WKH ORVV YHFWRU t LV JLYHQ 3HUP/HDUQ3UHF XSGDWHV WKH ZHLJKW YHFWRU pt LQ DQ DGGLWLYH ZD\ DQG SURMHFWV LW RQWR WKH VHW RI OLQHDU FRQVWUDLQWV UHSUHVHQW LQJ SUHFHGHQFH FRQVWUDLQWV 3UHFRQV A
(73) DQG WKH LQWHUVHFWLRQ RI WKH SHUPXWDKHGURQ Pn DQG 3UHFRQV A
(74) VXFFHVVLYHO\ 7KH PDLQ VWUXFWXUH RI RXU DOJRULWKP LWVHOI LV EXLOW RQ D VWDQGDUG RQOLQH FRQYH[ RSWLPL]DWLRQ DOJRULWKP 2QOLQH *UDGLHQW 'HVFHQW. F ,QIRUPDWLRQ 3URFHVVLQJ 6RFLHW\ RI -DSDQ . DQG ||q − pt+ || ≥ ||q − pt+ || + ||pt+ − pt+ || 6R E\ FRPELQLQJ WKHVH ZH KDYH ||q − pt || − ||q − pt+ || ≥||q − pt || − ||q − pt+ || + ||pt+ − pt+ || + ||pt+ − pt+ || ≥||q − pt || − ||q − pt+ || ,.
(75). ZKHUH WKH ODVW LQHTXDOLW\ IROORZV VLQFH (XFOLGHDQ GLVWDQFH LV QRQ QHJDWLYH 7KHQ E\ WKH IDFW WKDW pt+ = pt − ηt WKH ULJKW KDQG VLGH RI LQHTXDOLW\
(76) LV ||q − pt || − ||q − pt+ || = η q − pt
(77) · t − η ||t || . %\ FRPELQLQJ
(78)
(79) ZH FRPSOHWH WKH SURRI.
(80) . /HPPD &I =LQNHYLFK >@
(81) For any T ≥ and η = √ n +
(82) / T
(83) , 2.
(84) Vol.2014-AL-150 No.9 2014/11/20. ,36- 6,* 7HFKQLFDO 5HSRUW. T . pt · t ≤. PLQ. p∈Pn ∩3UHFRQV A
(85). t=. T . p · t +. t=. n n +
(86) √ T. . Proof. %\ /HPPD VXPPLQJ WKH LQHTXDOLW\ XS IURP t = WR T DQG DUUDQJLQJ WKHP ZH JHW WKDW IRU DQ\ q ∈ Pn ∩ 3UHFRQV A
(87) T T T η pt − q
(88) · t ≤ ||q − pt || − ||q − pt+ ||
(89) + ||t || η t= t= t=. η ||q − p || − ||q − pT ||
(90) + ||t || η t= T. =. n+ η n
(91) + nT ≤ η ZKHUH WKH ODVW LQHTXDOLW\ KROGV VLQFH IRU DQ\ i ∈ >n@ qi − pi,
(92) LV DW √ n PRVW p,i = n+
(93) DQG t ∈ >, @ %\ VHWWLQJ η = n +
(94) / T
(95) ZH KDYH WKH FXPXODWLYH ORVV ERXQG DV GHVLUHG . (ffiFLHQW ,PSOHPHQWDWLRQV RI 3URMHFWLRQ DQG 5RXQGLQJ ,Q WKLV VHFWLRQ ZH SURSRVH HffiFLHQW DOJRULWKPV IRU VXFFHVVLYH SURMHFWLRQV RQWR 3UHFRQV A
(96) DQG Pn ∩3UHFRQV A
(97) 7KHQ ZH VKRZ DQ LPSOHPHQWDWLRQ RI WKH SURFHGXUH 5RXQGLQJ 3URMHFWLRQ RQWR WKH 6HW 3UHFRQV A
(98) RI 3UHFHGHQFH &RQ VWUDLQWV 7KH SUREOHP RI SURMHFWLRQ RQWR 3UHFRQV A
(99) LV GHVFULEHG DV IRO ORZV PLQ ||p − q||. p∈Rn. VXEWR pi ≥ p j , IRU i, j
(100) ∈ A 7KLV SUREOHP LV NQRZQ DV WKH LVRWRQLF UHJUHVVLRQ SUREOHP >@ >@ >@ 3UHYLRXVO\ NQRZQ DOJRULWKPV IRU WKH LVRWRQLF UHJUHV VLRQ UXQ LQ O mn ORJ n
(101) RU O n
(102) WLPH VHH >@ >@ >@ IRU GHWDLOV 3URMHFWLRQ RQWR P n ∩ 3UHFRQV A
(103) 1RZ ZH VKRZ DQ HffiFLHQW DOJRULWKP 3URMHFWLRQ IRU FRPSXWLQJ WKH SURMHFWLRQ RQWR WKH LQWHUVHFWLRQ RI WKH SHUPXWDKHGURQ Pn DQG WKH VHW 3UHFRQV A
(104) RI SUHFHGHQFH FRQVWUDLQWV ,Q IDFW ZH ZLOO VKRZ WKDW WKH SUREOHP FDQ EH UHGXFHG WR SURMHFWLRQ RQWR Pn RQO\ 6R ZH ZLOO MXVW XVH WKH DOJRULWKP RI 6XHKLUR HW DO >@ IRU ¿QGLQJ WKH SURMHFWLRQ RQWR Pn )RUPDOO\ WKH SUREOHP LV VWDWHG DV IROORZV PLQ ||p − q||. p∈Rn. VXE WR . . pj ≤. |S | n + − j
(105) , IRU DQ\ S ⊂ >n@,. j∈S. j=. n . n n +
(106) , pj = . j=. pi ≥ p j , IRU i, j
(107) ∈ A )RU VLPSOLFLW\ ZH DVVXPH WKDW HOHPHQWV LQ q DUH VRUWHG LQ GH VFHQGLQJ RUGHU LH q ≥ q ≥ · · · ≥ qn 7KLV FDQ EH DFKLHYHG LQ WLPH O n ORJ n
(108) E\ VRUWLQJ q )LUVW ZH VKRZ WKDW WKLV SURMHFWLRQ. F ,QIRUPDWLRQ 3URFHVVLQJ 6RFLHW\ RI -DSDQ . SUHVHUYHV WKH RUGHU LQ q /HPPD 2UGHU 3UHVHUYLQJ /HPPD 6XHKLUR HW DO>@
(109)
(110) Let p∗ be the projection of q s.t. q ≥ q ≥ ... ≥ qn and A is the set of violating constraints w.r.t. q. Then the projection p∗ satisfies that p∗ ≥ p∗ ≥ p∗n . )XUWKHU ZH VKRZ WKDW WKH SURMHFWLRQ RQWR Pn SUHVHUYHV HTXDOLW\ DV ZHOO /HPPD (TXDOLW\ 3UHVHUYLQJ /HPPD
(111) Let p∗ be the projection of q. Then the projection p∗ satisfies that pi = p j if qi = q j . Proof. $VVXPH WKDW WKH OHPPD LV IDOVH 7KHQ WKHUH H[LVWV D SDLU i DQG j VXFK WKDW qi = q j DQG p∗i < p∗j /HW p EH WKH YHFWRU REWDLQHG E\ OHWWLQJ p i = p j = p∗i + p∗j
(112) / DQG p k = p∗k IRU k i, j ,W FDQ EH HDVLO\ YHUL¿HG WKDW p s ∈ Pn 1RZ REVHUYH WKDW ||p∗ − q|| − ||p − q||. ∗ . ∗ =p∗ i + p j − pi − p j + p · q − p · q ∗ ∗ ∗ . ∗. ∗ =p∗ i + p j − pi + p j
(113) / + pi − pi
(114) qi + p j − p j
(115) q j. = p∗i − p∗j
(116) + p i + p j − p∗i − p∗j
(117) qi ∗ = pi − p∗j
(118) > , ZKLFK FRQWUDGLFWV WKH IDFW WKDW p∗ LV WKH SURMHFWLRQ. . 1RZ ZH DUH UHDG\ WR VKRZ RQH RI RXU PDLQ WHFKQLFDO OHPPDV /HPPD For any q ∈ 3UHFRQV A
(119) , DUJ PLQ ||p − q|| = DUJ p∈Pn. PLQ. p∈Pn ∩3UHFRQV A
(120). ||p − q||.. Proof. /HW p∗ = DUJ PLQp∈Pn ||p − q|| %\ GH¿QLWLRQ RI WKH SUR MHFWLRQ IRU DQ\ p ∈ Pn ∩ 3UHFRQV A
(121) ⊆ Pn ||p − q|| ≥ ||p∗ − q|| )XUWKHU E\ /HPPD DQG p∗ SUHVHUYHV WKH RUGHU DQG HTXDOLW\ LQ q 7KDW LV p∗ DOVR VDWLV¿HV WKH FRQVWUDLQWV GH¿QHG E\ 3UHFRQV A
(122) 7KHUHIRUH ZH KDYH p∗ ∈ 3UHFRQV A
(123) 7KHVH IDFW LPSOLHV WKDW p∗ LV LQGHHG WKH SURMHFWLRQ RI q RQWR Pn ∩ 3UHFRQV A
(124) 6R E\ /HPPD ZKHQ D YHFWRU q ∈ 3UHFRQV A
(125) LV JLYHQ ZH FDQ FRPSXWH WKH SURMHFWLRQ RI q RQWR Pn ∩3UHFRQV A
(126) E\ FRPSXW LQJ WKH SURMHFWLRQ RI q RQWR Pn RQO\ %\ DSSO\LQJ WKH SURMHFWLRQ DOJRULWKP RI 6XHKLUR HW DO >@ IRU WKH EDVH SRO\KHGURQ ZKLFK JHQHUDOL]HV WKH SHUPXWDKHGURQ
(127) ZH REWDLQ WKH IROORZLQJ UHVXOW 7KHRUHP There exists an algorithm, with input q ∈ 3UHFRQV A
(128) , outputs the projection of q onto Pn ∩ 3UHFRQV A
(129) in time O n
(130) and space O n
(131) . 5RXQGLQJ :H VKRZ DQ DOJRULWKP IRU 5RXQGLQJ LQ $OJRULWKP 7KH DO JRULWKP LV VLPSOH 5RXJKO\ VSHDNLQJ LI WKH LQSXW p ∈ Pn ∩ 3UHFRQV A
(132) LV VRUWHG DV p ≥ · · · ≥ pn WKH DOJRULWKP RXWSXWV σ VXFK WKDW σ ≥ · · · ≥ σn LH σ = n, n − , . . . ,
(133) 1RWH WKDW ZH QHHG WR EUHDN WLHV LQ p WR FRQVWUXFW σ /HW A∗ EH WKH WUDQ VLWLYH FORVXUH RI A 6R JLYHQ DQ HTXLYDOHQFH VHW { j | pi = p j } ZH EUHDN WLHV VR WKDW LI i, j
(134) ∈ A∗ σi ≥ σ j 7KLV FDQ EH GRQH E\ HJ TXLFNVRUW )LUVW ZH VKRZ WKDW WKH URXQGLQJ JXDUDQWHHV WKDW σ ≤ − / n +
(135)
(136) 7KHQ ZH GLVFXVV WLPH FRPSOH[LW\ RI 5RXQGLQJ :H SURYH WKH IROORZLQJ OHPPD RQ 5RXQGLQJ /HPPD For any p ∈ Pn ∩ 3UHFRQV A
(137) s.t. p ≥ · · · ≥ pn , 3.
(138) Vol.2014-AL-150 No.9 2014/11/20. ,36- 6,* 7HFKQLFDO 5HSRUW. $OJRULWKP 5RXQGLQJ ,QSXW p ∈ Pn ∩ 3UHFRQV A
(139) VDWLVI\LQJ WKDW p ≥ p ≥ · · · ≥ pn DQG WKH WUDQVLWLYH FORVXUH A∗ RI A 2XWSXW 3HUPXWDWLRQ σ ∈ S n ∩ 3UHFRQV A
(140)
(141) 6RUW HOHPHQWV RI p LQ WKH GHVFHQGLQJ RUGHU ZKHUH IRU HOHPHQWV i, j VXFK WKDW pi = p j i LV ODUJHU WKDQ j LI i, j
(142) ∈ A∗ RWKHUZLVH EHDN WKH WLH DUELWUDULO\
(143) 2XWSXW WKH SHUPXWDWLRQ σ VW σi = n +
(144) − ri ZKHUH ri LV WKH RUGLQDO RI i LQ WKH DERYH RUGHU. the output σ of Rounding given p satisfies that for any i ∈ >n@, σi ≤ − / n +
(145)
(146) pi . Proof. )RU DQ\ i ∈ >n@ E\ GH¿QLWLRQ RI WKH SHUPXWDKHGURQ ZH KDYH i . pj ≤. j=. i− . j=. j=. i i −
(147) . . n. j=i. j. n+−i. n =. j=. j−. i−. j=. n+−i. j. ≥. n + i
(148) n + − i
(149) n + i = . n + − i
(150) . $OVR VLQFH WKH URXQGLQJ DOJRULWKP RXWSXWV σ VW σi DFFRUGLQJ WR GHFUHDVLQJ RUGHU RI pi ZH KDYH WKDW σ ≥ σ ≥ · · · ≥ σn DQG σi = n − i + IRU i = , , . . . , n 7KXV IRU DQ\ i ∈ >n@ i − n − i + n + i −
(151) σi =− . = ≤ pi n + i n+i n + i
(152) +HUH WKH VHFRQG WHUP i− n+i LV PLQLPL]HG ZKHQ i = 7KHUHIRUH σi /pi ≤ − / n +
(153) DV FODLPHG )RU FRPSXWLQJ 5RXQGLQJ ZH QHHG WR FRQVWUXFW WKH WUDQVLWLYH FORVXUH A∗ RI A EHIRUH WKH SURWRFRO VWDUWV ,W LV ZHOO NQRZQ WKDW D WUDQVLWLYH FORVXUH FDQ EH FRPSXWHG E\ XVLQJ DOJRULWKPV IRU DOO SDLUV VKRUWHVW SDWKHV )RU WKLV SUREOHP )OR\G:DUVKDOO DOJRULWKP FDQ EH XVHG DQG LW UXQV LQ WLPH O n
(154) DQG VSDFH O n
(155) VHH HJ >@
(156) :KHQ A LV VPDOO IRU H[DPSOH m << n ZH FDQ XVH -RKQ VRQ¶V DOJRULWKP UXQQLQJ LQ WLPH O n ORJ n+nm
(157) DQG VSDFH O m
(158) 7KH WLPH FRPSOH[LW\ RI 5RXQGLQJ LV O n
(159) ZKLFK LV GXH WR WKH VRUWLQJ 7KH VSDFH FRPSOH[LW\ LV O n
(160) LI ZH XVH )OR\G:DUVKDOO DOJRULWKP ZLWK D DGMDFHQF\ PDWUL[ 7KH VSDFH FRPSOH[LW\ FDQ EH UHGXFHG WR O m
(161) LI ZH HPSOR\ -RKQVRQ¶V DOJRULWKP ZKLFK XVHV DQ DGMDFHQF\ OLVW 2Q WKH RWKHU KDQG ZH QHHG DQ H[WUD O ORJ m
(162) IDFWRU LQ WKH WLPH FRPSOH[LW\ VLQFH ZH QHHG O ORJ m
(163) WLPH WR FKHFN LI i, j
(164) ∈ A∗ ZKHQ A∗ LV JLYHQ DV DQ DGMDFHQF\ OLVW 0DLQ 5HVXOW 1RZ ZH DUH UHDG\ WR SURYH WKH PDLQ UHVXOW %\ /HPPD DQG 7KHRUHP ZH JHW WKH IROORZLQJ WKHRUHP LPPHGLDWHO\ 7KHRUHP There exists an online scheduling algorithm over Pn ∩ 3UHFRQV A
(165) such that √ ( 1 ) its − / n +
(166)
(167) -regret is O n T
(168) , and ( 2 ) its running time is O n
(169) time per trial.. F ,QIRUPDWLRQ 3URFHVVLQJ 6RFLHW\ RI -DSDQ . ,Q WKLV VHFWLRQ ZH GHULYH D ORZHU ERXQG RI WKH UHJUHW IRU RXU RQOLQH VFKHGXOLQJ SUREOHP RYHU WKH SHUPXWDKHGURQ Pn +HUH ZH FRQVLGHU WKH VSHFLDO FDVH ZKHUH QR SUHFHGHQFH FRQVWUDLQW LV JLYHQ 7KHRUHP For our online scheduling problem over the permu√ tahedron Pn , for sufficiently large T , the -regret is Ω n T
(170) . Proof. :H FRQVLGHU WKH DGYHUVDU\ ZKLFK SOD\V UDQGRPO\ 0RUH SUHFLVHO\ DW HDFK WULDO t WKH DGYHUVDU\ FKRRVHV D ORVV YHFWRU t UDQGRPO\ IURP , ZKHUH
(171) EH WKH ORVV YHFWRU LQ ZKLFK WKH ¿UVW n HOHPHQWV DUH V V
(172) DQG WKH UHPDLQLQJ HOHPHQWV DUH V V
(173) 7KHQ IRU DQ\ RQOLQH VFKHGXOLQJ DOJRULWKP ZKLFK RXWSXWV σt ∈ S t DW WULDO t T n n +
(174) T . E> σt · t @ = t=.
(175). %\ WKH DVVXPSWLRQ WKDW p ≥ · · · ≥ pn WKH DYHUDJH RI pi + pi+ + · · · + pn LV QRW ODUJHU WKDQ pi 7KXV ZH JHW WKDW pi ≥. /RZHU %RXQG. 1RZ OHW XV FRQVLGHU WKH EHVW ¿[HG SHUPXWDWLRQ /HW σ = n, n − , n − , . . . ,
(176) DQG σ = , , , , . . . , n
(177) UHVSHFWLYHO\ 6XSSRVH WKDW DSSHDUV PRUH IUHTXHQWO\ WKDQ E\ k 7KHQ WKH EHVW SHUPXWDWLRQ LV σ DQG LWV FXPXODWLYH ORVV LV n n T k T k + − i i + i= i= n + n n n n +
(178) T k +
(179) n n +
(180) k + − = n + n n +
(181) T k n + − n − = . =. n n +
(182) T k n − . . 7KH VDPH DUJXPHQW IROORZV WKH RSSRVLWH FDVH ZKHUH LV PRUH IUHTXHQW E\ k ,Q IDFW k FDQ EH H[SUHVVHG DV k = Tt= δt ZKHUH HDFK δt LV D UDQGRP YDULDEOH ZKLFK WDNHV YDOXHV ± HTXDOO\ UDQ GRPO\ 7KHQ WKH H[SHFWHG UHJUHW RI DQ\ RQOLQH VFKHGXOLQJ DO JRULWKPV LV DW OHDVW n E Tt= δt %\ WKH FHQWUDO OLPLW WKHR UHP WKH GLVWULEXWLRQ RI Tt= δt FRQYHUJHV WR *DXVVLDQ GLVWULEX √ WLRQ ZLWK PHDQ DQG YDULDQFH T 6R IRU VXffiFLHQWO\ ODUJH T √ 3U>| Tt= δt | ≥ T @ LV FRQVWDQW VD\ c < c <
(183) 7KHUHIRUH √ WKH H[SHFWHG UHJUHW ERXQG LV IXUWKHU ORZHU ERXQGHG DV n c T 7KLV LPSOLHV WKDW WKHUH H[LVWV D VHTXHQFH RI ORVV YHFWRUV WKDW HQ IRUFHV DQ\ RQOLQH VFKHGXOLQJ DOJRULWKP WR LQFXU UHJUHW LV DW OHDVW √ Ω n T
(184) ,Q IDFW WKLV ORZHU ERXQG RQ UHJUHW LV WLJKW LQ JHQHUDO VLQFH √ WKHUH DUH RQOLQH DOJRULWKPV ZKLFK DFKLHYH UHJUHW O n T
(185) >@ >@
(186) √ 1RZ LW LV QDWXUDO WR DVN LI WKH − / n +
(187)
(188) UHJUHW O n T
(189) LV WLJKW XQGHU SUHFHGHQFH FRQVWUDLQWV 6R IDU ZH KDYH QR ORZHU ERXQG IRU WKLV FDVH %XW ZH JLYH DQ DOWHUQDWLYH DUJXPHQW WKDW RXU DOJRULWKP LV RSWLPDO XQOHVV WKHUH DUH DQ RfflLQH DOJRULWKP ZLWK DSSUR[LPDWLRQ UDWLR α < 7KHRUHP If there exists a polynomial time online scheduling √ algorithm with α-regret poly n, m
(190) T , then there also exists a randomized polynomial time algorithm for the offline problem with approximation ratio α.. 4.
(191) Vol.2014-AL-150 No.9 2014/11/20. ,36- 6,* 7HFKQLFDO 5HSRUW. Proof. 7KH SURRI LV EDVHG RQ VWDQGDUG RQOLQH WR RfflLQH FRQYHU VLRQ PHWKRGV LQ WKH RQOLQH OHDUQLQJ OLWHUDWXUH VHHHJ> @
(192) /HW A EH VXFK DQ RQOLQH VFKHGXOLQJ DOJRULWKP DQG LWV RXWSXW DW HDFK WULDO t LV GHQRWHG DV σt /HW ∈ >, @n EH WKH ORVV YHFWRU LQ WKH RfflLQH SUREOHP :H FRQVLGHU WKH DGYHUVDU\ ZKLFK UHWXUQV t = DW HDFK WULDO t 7KHQ WKH WKH FXPXODWLYH ORVV RI A GLYLGHG E\ T LV ERXQGHG DV T poly n, m
(193) . σt · ≤ α PLQ σ·+ σ∈S n ∩3UHFRQV A
(194) T t= T. EH D XQLIRUPO\ DQG UDQGRPO\ FKRVHQ SHUPXWDWLRQ 1RZ OHW σ IURP {σ , . . . , σT } 7KHQ E> σ · @ ≤ α. PLQ. σ∈S n ∩3UHFRQV A
(195). σ·+. poly n, m
(196) . T. LV %\ VHWWLQJ T = poly n, m
(197) WKH H[SHFWHG FXPXODWLYH ORVV RI σ DW PRVW α WLPHV WKH FXPXODWLYH ORVV RI WKH EHVW SHUPXWDWLRQ ZLWK D FRQVWDQW DGGLWLYH WHUP
(198) ZKLFK FRPSOHWHV WKH SURRI . &RQFOXVLRQ DQG 2SHQ 3UREOHPV ,Q WKLV SDSHU ZH SURSRVH D SRO\QRPLDO WLPH RQOLQH MRE VFKHGXO LQJ DOJRULWKP RYHU 3HUPXWDKHGURQ XQGHU SUHFHGHQFH FRQVWUDLQWV √ 2XU DOJRULWKP DFKLHYHV − O n
(199)
(200) UHJUHW O n T
(201) ZKLFK PHDQV WKDW RXUV FDQ SUHGLFW DV ZHOO DV WKH VWDWHRIWKH DUW RfflLQH DSSUR[ LPDWLRQ DOJRULWKPV LQ KLQGVLJKW ,Q IDFW WKH UDWLR α = − O n
(202) LV WLJKW LI WKHUH LV QR RfflLQH DSSUR[LPDWLRQ DOJRULWKP ZKRVH DS SUR[LPDWLRQ UDWLR LV VWULFWO\ OHVV WKDQ − / n +
(203) $Q LQWHUHVWLQJ RSHQ TXHVWLRQ LV H[WHQGLQJ RXU RQOLQH IUDPH ZRUN IRU PLQLPL]LQJ WKH VXP RI ZHLJKWHG FRPSOHWLRQ WLPHV. >@ >@ >@ >@ >@. >@ > @. >@ >@ >@ >@. >@ >@. 6 .DNDGH $ 7 .DODL DQG / /LJHWW 3OD\LQJ JDPHV ZLWK DSSUR[ LPDWLRQ DOJRULWKPV SIAM Journal on Computing
(204) ± ( / /DZOHU 2Q 6HTXHQFLQJ MREV WR PLQLPL]H ZHLJKWHG FRPSOHWLRQ WLPH VXEMHFW WR SUHFHGHQFH FRQVWUDLQWV Annals of Discrete Mathematics 2 ± - . /HQVWUD DQG $ + * 5 .DQ &RPSOH[LW\ RI VFKHGXOLQJ XQGHU SUHFHGHQFH FRQVWUDLQWV Operations Research ± 5 /XVV 6 5RVVHW DQG 0 6KDKDU (ffiFLHQW UHJXODUL]HG LVRWRQLF UH JUHVVLRQ ZLWK DSSOLFDWLRQ WR JHQHJHQH LQWHUDFWLRQ VHDUFK Annals of Applied Statistics
(205) ) 0DUJRW 0 4XH\UDQQH DQG < :DQJ 'HFRPSRVLWLRQV 1HWZRUN )ORZV DQG D 3UHFHGHQFH &RQVWUDLQHG 6LQJOH0DFKLQH 6FKHGXOLQJ 3UREOHP'HFRPSRVLWLRQV 1HWZRUN )ORZV DQG D 3UHFHGHQFH &RQ VWUDLQHG 6LQJOH0DFKLQH 6FKHGXOLQJ 3UREOHP Operations Research
(206) ± : 0D[ZHOO DQG - 0XFNVWDGW (VWDEOLVKLQJ FRQVLVWHQW DQG UHDOLVWLF UHRUGHU LQWHUYDOV LQ SURGXFWLRQGLVWULEXWLRQ V\VWHPV Operations Research ± $ 6 6FKXO] 6FKHGXOLQJ WR 0LQLPL]H 7RWDO :HLJKWHG &RPSOHWLRQ 7LPH 3HUIRUPDQFH *XDUDQWHHV RI /3%DVHG +HXULVWLFV DQG /RZHU %RXQGV ,Q Proceedings of the 5th Conference on Integer Programming and Combinatorial Optimization (IPCO1996) SDJHV ± - 6SRXJH + :DQ DQG : :LOEXU /HDVW VTXDUHV LVRWRQLF UHJUHVVLRQ LQ WZR GLPHQVLRQV J. Optimization Theory and Apps. ± ' 6XHKLUR . +DWDQR 6 .LMLPD ( 7DNLPRWR DQG 1 .L\RKLWR 2Q OLQH SUHGLFWLRQ XQGHU VXEPRGXODU FRQVWUDLQWV ,Q In Proceedings of 23th Annual Conference on Learning Theory (ALT 2012) * - :RHJLQJHU DQG 3 6FKXXUPDQ 3RO\QRPLDO WLPH DSSUR[LPDWLRQ DOJRULWKPV IRU PDFKLQH VFKHGXOLQJ 7HQ RSHQ SUREOHPV Journal of Scheduling ± 6 <DVXWDNH . +DWDQR 6 .LMLPD ( 7DNLPRWR DQG 0 7DNHGD 2QOLQH /LQHDU 2SWLPL]DWLRQ RYHU 3HUPXWDWLRQV ,Q Proceedings of the 22nd International Symposium on Algorithms and Computation (ISAAC 2011) SDJHV ± * 0 =LHJOHU Lectures on Polytopes *UDGXDWH 7H[WV LQ 0DWKHPDWLFV 6SULQJHU9HUODJ 0 =LQNHYLFK 2QOLQH FRQYH[ SURJUDPPLQJ DQG JHQHUDOL]HG LQ¿QLWHV LPDO JUDGLHQW DVFHQW ,Q Proceedings of the Twentieth International Conference on Machine Learning (ICML ’03) SDJHV ± . 5HIHUHQFHV >@. >@. >@. >@ >@ >@ >@ >@ > @ >@ >@ >@. 1 $LORQ ,PSURYHG ERXQGV IRU RQOLQH OHDUQLQJ RYHU WKH SHUPXWDKH GURQ DQG RWKHU UDQNLQJ SRO\WRSHV ,Q Proceedings of 17th International Conference on Artificial Intelligence and Statistics (AISTAT2014) SDJHV ± & $PEXKO 0 0DVWUROLOOL 1 0XWVDQDV DQG 2 6YHQVVRQ 2Q WKH $SSUR[LPDELOLW\ RI 6LQJOH0DFKLQH 6FKHGXOLQJ ZLWK 3UHFHGHQFH &RQ VWUDLQWV &KULVWRSK $PEXKO Mathematics of Operations Research
(207) ± / 0 %UHJPDQ 7KH UHOD[DWLRQ PHWKRG RI ¿QGLQJ WKH FRPPRQ SRLQW RI FRQYH[ VHWV DQG LWV DSSOLFDWLRQ WR WKH VROXWLRQ RI SUREOHPV LQ FRQ YH[ SURJUDPPLQJ USSR Computational Mathematics and Physics ± 1 &HVD%LDQFKL DQG * /XJRVL Prediction, Learning, and Games &DPEULGJH 8QLYHUVLW\ 3UHVV & &KHNXUL DQG 5 0RWZDQL 3UHFHGHQFH FRQVWUDLQHG VFKHGXOLQJ WR PLQLPL]H VXP RI ZHLJKWHG FRPSOHWLRQ WLPHV RQ D VLQJOH PDFKLQH Discrete Applied Mathematics
(208) ± ) $ &KXGDN DQG ' 6 +RFKEDXP $ KDOILQWHJUDO OLQHDU SURJUDP PLQJ UHOD[DWLRQ IRU VFKHGXOLQJ SUHFHGHQFHFRQVWUDLQHG MREV RQ D VLQ JOH PDFKLQH Operations Research Letters ± 7 + &RUPHQ & ( /HLVHUVRQ 5 / 5LYHVW DQG 6 &OLffRUG Introduction to Algorithms 7KH 0,7 3UHVV WKLUG HGLW HGLWLRQ - 5 &RUUHD DQG $ 6 6FKXO] 1R 7LWOH Mathematics of Operations Research
(209) ± < )UHXQG DQG 5 ( 6FKDSLUH /DUJH 0DUJLQ &ODVVL¿FDWLRQ 8VLQJ WKH 3HUFHSWURQ $OJRULWKP Machine Learning
(210) ± 6 )XMLVKLJH Submodular functions and optimization (OVHYLHU 6FL HQFH QG HGLWLRQ 7 )XMLWD . +DWDQR DQG ( 7DNLPRWR &RPELQDWRULDO 2QOLQH 3UHGLF WLRQ YLD 0HWDURXQGLQJ ,Q Proceedings of 24th Annual Conference on Algorithmic Learning Theory (ALT 2013) SDJHV ± / $ +DOO $ 6 6FKXO] ' % 6KPR\V DQG - :HLQ 6FKHGXOLQJ WR 0LQLPL]H $YHUDJH &RPSOHWLRQ 7LPH 2ff/LQH DQG 2Q/LQH $SSUR[L PDWLRQ $OJRULWKPV Mathematics of Operations Research
(211) ± . F ,QIRUPDWLRQ 3URFHVVLQJ 6RFLHW\ RI -DSDQ . 5.
(212)
関連したドキュメント
Hence, the degree theory constructed in [1] is an extension of the classical Leray–Schauder degree in Hilbert space.. It is unique, single-valued and has the usual properties of
She reviews the status of a number of interrelated problems on diameters of graphs, including: (i) degree/diameter problem, (ii) order/degree problem, (iii) given n, D, D 0 ,
We study the description of torsion free sheaves on X in terms of vector bundles with an additional structure on e X which was introduced by Seshadri.. Keywords: torsion-free
If in the infinite dimensional case we have a family of holomorphic mappings which satisfies in some sense an approximate semigroup property (see Definition 1), and converges to
(The Elliott-Halberstam conjecture does allow one to take B = 2 in (1.39), and therefore leads to small improve- ments in Huxley’s results, which for r ≥ 2 are weaker than the result
のようにすべきだと考えていますか。 やっと開通します。長野、太田地区方面
In this section we consider the submodular flow problem, the independent flow problem and the polymatroidal flow problem, which we call neoflow problems.. We discuss the equivalence
For example, it is not obvious at all that the invariants of rooted trees given by coefficients of the generating functions f (t ), ˜ d(t ), ˜ h(t ) ˜ and m(t ) can be obtained
