. . .. .I. ......‘0 9‘3 .‘t . . . . .. . . . .8403? ..A...‘o... 0 3...... ....n... ......J... . . . . ......r... “v.31.-. . I 1.11.411 . 3 . .. . .. . . . -..... . .-....z..fl1..........u......a...... 3...... 1................. 2.....- ...... .... .. .... .n... ......fi...3...... .. .. .. - I u.. I v .. . . ... .4 a c Q . . . . . . t..fljfi1lu.2.flt‘.&.|2'tuito. ...!oxu. 30.2.... 3 ...-... .12! .2. 5'3... 43.9.0”. 830. z... I .2 . 1.9.36.- . ......DQt ... . . 1.- . '1: .0. ......- a. 0.. ... 5' 9!.3.0’.Q1....v9"(00.. 03100$00| .... .. ... .00 .I I 11.1.... (2973“. ‘1’.F)!!$fi.bfl'dmt!$l*l.€83.'9 aims-.2030... x...’l(t7i ......(2..r..£.o.am.:l.u’ . .- ... O ...]: :17.» 5.... t t I. Elva} 3......Io‘lo.:..t..ficr..8 9316.335... ... 23.029... 3...... ... 01.... ..p ......o '8 . 0..“ifi7‘918! ...-opts. . 3 o ...-61. .....o. IQ...'6.C~ 07.13.330.39!!! .....MI..‘...! I co¥.l.0 ...o. u .0 f 0.... . ....‘k.".. .5... '..‘-.~..‘.. 3 . .33.. . g“ C... I... 0. .0... L: k I). ... 0...... ...-0. Gonlv. .i u 0.... .. I. 300.39-). (343.0!“ .....rrxflnuvar. £20320. .. ”tuna-om 6.3.... .. .... v.7. 33-25.10. gritty. 30.3...3831"! . 0C .. OJ? ...-.1...” 00.ka . . O . : .- ......3..a 0.. . .... 0...C..0.£C . . 0.81% ...-......H... I: ... ...-0.. o 0.9..o...to.a. ...-£0... . 1'32, ... ... D I ‘4. I ob. ... l “V.- .1 $0": I... . o X .00. ... .. ‘0‘ ... . . O. u «8...... {00. ...mhz . or: 0’¥"!J..2.’.dalifu.ouullfz.’2 .l‘?" ... .... . . . . ...“denwlizlllkuLfl .7... .SJJ.’ 5.3)..) 2 I .3", . . ’10 0. :II’... #:3003321. O.....0.z.. 601......3 i... .. ... ....9. .... . . . . .. 2!... 2.3.13.7. o ... ...... ... . .....«O: .0...pl> 9 ‘ I“ no... .0 . ...“ OI. .00..ev1’.¢0 0.0V ‘00. 0.0... . .0... 3'... 30.0 51".3sop- . .,. 3K}: . , . .. . . . . . ”I‘I. Ital. ...f. y‘.‘ .51. ... ”fil";.1a 3"... 0.... {tit-$133.. . l._ r... r.- In". ‘90—'81: 0.": .1330. I .1170 0!... ....I... ...-lrtaiat. . . . . . . “...“. . ...".OISD . 40.1 v ...1 . 8 I .’ o 03". it‘d:‘ ’Ivallo....‘ .a ...1..3"'.0.o: ...... 0:... .... '0... . #O. ‘3 ...... 3D.‘}..I......D ..|'. .. s. . .0 I ...“...u‘. 3." .12 .0 .... .3! .0. 9 0003.09 . 9.Lou0..f..;’oozr9o 00%....o...‘ atomic-... .ei..1!...a..u.!..10.:<. l0. V V l .. . I. 13...! ......" .... ‘30. ....’..3v“1av0.3..lr...looo.orf.. ....0...??..0. I0‘......00...0. 3.... 0... .-.-58:00.20... . ...-ll... VI“ 0;... 800......ng‘0... .0......:lr.a§i..l.l$982.9.5133... . agar... r 3.21.0933. .. .43“: 245.5...“ ...“! ...et... “I’vzzfisvfluu uh. ... st... . 0.. . .... I. I: .... . VJ. . .0. .. 3"..l- . ..o .‘ .... l 0'. “MW...;Y. d; ......I ..0. . 1...... J. .- ... a. .' ...... . 1.08 . At. . . . .0... ’7‘. 7'”... . ...... .. .. ol. . .. ... gold... fifiuo l..‘..1|JNA-COO.1....H;£- .04.. . 3 to... .2. . 0" . 0". . . t p ' ‘0!" s . . ow unaffi‘ 69“.. 'Jzu ‘uaaoofn v}§thu¥n.’?|‘ .0 2 . . .nof . .0. . .... l .0. H". .4... ...«o’ ...V. ... 0...", ....) ... 0.. I . .. - 2.0.... a... ......hflgmwr C “"1220. .r..._...o.2¥.....;23.‘.l .ow . ~ . . 9 .2! . .13... 3.3.4.... . ....rs’ . . . £01.33... . ...... IP'...0.. .9 .. .1... 0'09}. 0.15.0. .... ....‘Oo. ... u" . . . . o 0 . .l . ’00-! . ... vnQu-d.’ “Gun“.- .00.... o. .3". ‘3‘... ‘l. 0.... ‘ m I3? ...... . .o h .. 61.15.... . I 1. [...-3 I ...0“O¢...0.0‘.0“.S ‘u‘ .00. . . 9 to. ...-1.0. . ... ... . ..“......II.. . l.’ 0.. V v 0'.‘I!.C.... . J0ci‘f’i. “04......“v.0".‘fl'uaflfl193233'3'll1...”..03...’ ....sfi in a 19.4... ..:.Our,$v.'v .01.. ‘33 ......II. ..0! . 3...». Pe..'..x .. .. ..Iu.....3?’..:! l .... .- ... 3...: . it"s...) ...... I is; . 0.0 ’fi)‘.zfi u... ' l0 .... ...}.k. .. ...Qu..¢ . 2.3.... $00.00.: I ... . ... f0.f‘o..0'.0bvo.ouuou.v.2:. . 8|‘a0'l'..".c.00...23:8 9...: ...:x..t!00...§0t70..£. . ....J I fiQC ...-lilr..noo.t ...-1...... 101......92 8 90.01.33... . ....._..0.0r...\.. .0...:(:..’ ... ... 31"... O’J...JI.."...¢IR1: . » 3.10.03... ...“.w..».ob 221.14... ,.”.v‘.3.1 .0. 2131...: p . . g. 04",...1 .. 7...... 1.‘ .l 6.. 6.0 a... 4‘ 3!: 0mm Bud"... r ......L...0..Q’.c '4. LC .... . . . . . 3.0.0... ...!vav... . . . 8‘103’8! Avon...‘.¢!oo‘¢- . . . . . . . .. .O‘QI 3.10.0.3? ...... . . . ......) r ... 348.037.”...;31noau.0+_.l..3:..l:h .0 20.... .... “nu? ...: 3. .. {Oil-v 216.0 ll!!......l.:..90l. .0. .- 0.) 3 .9) ...a . ... 1.30.9.1. .- 1. . W. M5. . ..J‘ .2332?! .... .(..l.‘_.. .I o.- .'. .71.)...» .....}!...,?.~&V..ID :9 l.‘ 1.0.5.... . . (....Jfi.: . 04 0w 1...... r ...-0... .1. 5"... In... 9... 9.91.1. .800 0-....'Q..II¢ 9 .0... _ . .-.... .... u. .... J .... if.» t03..l&0.. ...... .. :0. 0...... ..II...-O0.2‘...... 0.190.231"?! . . . . . . .... .. . ..Jlt..19..ol.0t..’ ... .100... 31‘..‘. .‘0.:\‘-.;2 o..il¢.0.:..v9..\‘. u...’.0{ £3,300. '30....3 . .... . ... .3 039. . ... .... ..o ‘.5 . {33' . . .... Ja ..l! ...... Ilgolu.O—h&v201 ....1023. 99'1:.Iv6.'. S {Ifdp . .... ..O‘ a. a . .5...- . . . . . . . .. . . .u... . r .....3...»... .... ... 30.25.1133: . . : ...rl 8:52.381». 39...... 1.....2 .... r .0 . .. . 502:2?! ...... o!!......§t.: .‘e.1¢9::....ax... ... . . . . “......” a. ... : 20.. ...: 0. ..0.‘ 3...! .. if: .n...»...‘.. 3"...)3 1.353....0... '09.... ...I... ... 0......11033: ... 0...- J. '2... 30.. ... ...-«1-380.013.9242 ... . . . v . I ...... ...-:13?! . . .2...l.....l..r{..r..ao.. .2"... 33.323003: . 3...... .... .68.... 2.2.3....- ‘Qiltgxt. ...! I: ....6:)..itt...tnuvo.t:::!:lt.95:2... $0.3... ............I..:.. .9... .. .c.t.........:. ._ r. ...“..a... ...: : t ... )3...“ ..l........0..‘.fs .85.... .o. 8). 39.9.25. 10.000 m “be?! . .c ... 70 [...-.70.... . 31.019359.‘i......ttt$..£'.. .zu‘x...!t:.. 91.3.3.3. . 03.0... .3? .... 3......v..r1 . . . . ...... o (J. ... .. . ....u......10$’3. 3“” 7.0.2 0.. J .2533...“ §a¢!§::3!.0.§ .201.” a I. ....38..PC..’3}.1 ...-0. ... .15.. 2... ....3.....‘....\¢Io....o. ......02..;... 2...-:9. ......l...3.:0.0!. ...... . ...: .. t a. ..J. «til 0...... 1.7.1.0 ...!!‘dtiaoi’. it. 1.0.. . 0.1.9.211! 6.2.28. ...... 3... ......MXI... l. -.. 3.1.... 3.2.-.. 96.1.8.3. .....to... .1!.......... 55.3.1.3.1... ......z... 1 ...!(03 . “(:1 ...!!29 I . ...I ..P...1 .10. . .. ... v 3...... O.» to] ..\0£..0.3!3 2‘5...“ 34941.9... ...... .9. . 3'3... 088....-. ...-231....8301{fuzttofoottz .‘-...o.... .0... 1...... .3... .l .. -...riv . ¢ .. “ORV-0 VO..I.OIIL.~W . .. ...... 0.-.? .P .LWQP..P 0.... ...-.13. . ...Jt.‘).‘.oot.l8llt:l: .......1:v':......t 1.1.3.9....)83 ...!i . :0 ... . ...... ...... . ...}...Stn... 8.1.6231... ......5621. . .... .... .... a a o O ..Jo.l.. u“! .6 . 31...; . o: .00...) . 4.5.3.9.‘1'37. t...{.zo'.1‘.l.0'....o.. ...; .‘.+’lf£30r..3....‘.....0. - . 33.. 33.09 :32. .200. .. o.C.r.—\....c.3.oif0. .. 2.0.? . 93H.00O'10C.2.!’ .‘f .. ...”.V'... . .0 a. ...: 2... 319.13... 1...! ..."... .... z... 5...... N7... ...... ...... .... 12...... «4.14:...rsh... ......2...:..r . ......1... ...-..32 ......i: 31...... .12.... .... .... .. .. .... .... . . . . ... ......" .... . ...... «£334.. ...... _. 9... ...... 81:33.“?! ..rtlxp ... ...-... ...x..lod ...: ....9..... 32...... 0.. .... .75....- . 5... (...-...!!! . . 9...... .8 .. 2.5.6.... :03... ...... . . 01.33.... .3302... .-....ZC 4 3.3.3.1132 . . . . ..u' b.0093. ’99.. ... .3 ital-1.... I. IS . ....J 531.0%..0100: ....".llf.. .0 $30.63... P... ...§9 93...... 609.33.. r‘. .32}. . . 3...... c Q... . . It... 42:35:03... 1.3.0.! 1.3.3... ... .789...00.$1..2 . . D .Illli'qa A . ...... £51....) ... I. . 13.0,834 r)..t..4.....l.._.:3159. W. .....‘t I 1.. 1.1.0:]?....Ol..t..€3"2..:...0..... 3 322..-...139 .9 .....z... .lt.... ......o..._...l:v . £00.2{A29 ! £9 04¢... a r o . I .43:64336..€1¢. G ......i of. .. «o 30.! lo‘lavl... '9 2.010,; ... ....QI.._I.V. ...0.c...t. O’ccl... .0... ..I. . .. . . _ . . 2.32... a. ......h. . .91.- 4.‘ 12.3631... . .00.. ...} .u'..\. 0.3on v .052... 2..-0......02'..! 8......0...l .10.. .3... ‘70:..8... . \ . . . .. . . l.._....o.!. ‘58.???” ‘2... . . v0.0.3.3 . 0.... ...l ..Jf....v.il.....40:... ...? 00......Jaé‘tla . . . . .. . . . . _ . 06... £553.. ... ......p. ..0. F)... .. 3.4.0.331 v. ...-.8 I .33).... .33.. .301... 1..lop..lo..c! . .. t... ..3.?9.0 ..."..T .. .30»... ...” 21.3..ldaa .. 30»... ...-......t: Jtlzzkgé ...-.9... . ‘3’}...(J’ditt. o‘ ...oo' 10.4.... 509.. .790. ‘.’.~ ...-53....-.“ .3... 209.0}..3.6 .0 5.3.1.. ... 80., . .. ... ...i 1.138.! ....aia.‘u~l.v.(.‘ .0 031.0... 1 93.05 ‘33..."‘21‘. o3 0‘..:..0..AI..... v 1:99.. 34.20.... .. 19.8.. .. .Ptl... .. :83... 1.5.: . 1...... .... ....0: A «8.0.... u 1...... I... util‘o . so... ’ ‘30PQ*-. . . . . . 1 . z......:...... If; .0. . .. . . . .9. ...g’? 2.....50 lvk',anoom.\vfi‘mr§ . . . _ . . . . t ,....a..z.a...n“__......_.....o.i.. .. .....2: .. .. .. . . .. . . o . .. . . , . ...! 0"... 0.1. . ... . .. . 0920.0... . . . . . . . . . . . . ..'.Q...‘Vq.. . 00,..- . . . tinnilv. .... ...... 01H 3‘!!! . ...o.‘ x sat)...» ...-2..... .v . .. Jolt-......‘l .....X..o..t’ 9‘. 0.. 03.10.... W ...: 22.....nl . 3 839‘ . 178... I. . . 0%.”. ......logu' ...‘OJ. 0. nor I. .l-..‘.r.. .9 1.. . . . . . . .. .... .Nl..... 7...... ...: "1.1.x. 2.6... . . . . . . . . . . . ... . ... 9'. ...... M0503: '.‘t~w4.. ... . , .. . _. . . . 5.83.6.2... . .. 9.3.9.. :94th k. 9.2%... (20., On . . . . . . . . . . . : . 6.68.... 330.605... .91. .... 03,: 0: I4. '33.“... . Sis-0:5... .01 i .. . o .. . . . . . tit; ...‘K.’?'. ,.O.‘ t 0. In. “I3 g}. ‘10; ._ . U . .. ...OO...!... 1... 5:...l,l€l.vo.. . . . . . . . . . . . . . 8 .I' .... 30...... II... ”‘3: . ”Q .... 3.. .0... .1 0.5.2. .0!.....—‘J.\.a.uo..nu f... . v3.9.8... .. .. . . . . . .. . . . . . . . Z ... ”huh.....u.~u~..‘mkfl:ufl. 0.. . . ... - o v o 1.! . 0.x .. .0 O ‘0. on“: ..a. I‘D. 0.1.1., . . . I I... 900. .. . . . . . . . . . a . O! ...J.|d...a‘9.(3.u.. u..u.00.0.too? \ 535%.”; ”101.....34; Ono‘flluo. 0.02230 ...-Fl... 21.3.... . 0.... . O... .0 .... .... .v‘QO m2 4.0....20..$.J..|:. 01.. . 3.9.7.1....- fog.- .t‘lt‘ 0D..fl ....d .300... . . ..d... . ’rf..i.0.o.0 o . . :10. .. ._ . . . b’£.¢.’J\..tIO.L .05....2 .‘ . .2013: ......0. .... 2 81:1... .3..1&‘84{1.‘9v.. . . ’09 ...!O. 0.." . .... . ... .4... . ....I. 2.91... J... . .... ~ .41. ... ...... .01... .... o0 . .... .JQouu. . ’x. ... ....”u‘ ..3 .0. .. Jofl. : ... 129.26....JI' . 1:3...9. . z... 62...“... 0...?! (I... 00....Pn...u’.v‘0”.rdu.‘1.u.a.fl.w 0.90.. ‘20.. .. '3? ..D....;...o . . .... ...... ... $43!...- .... . 90 a. I...- . F v .. .... . . ...! ... .... . .. . ... ..t' I... .....- .....0. l .- .. .2. .... .1. ‘0 (.14 ..1 0.1.: 3. 0.1.00. .o‘.‘ ...}? .1 4'5 . . . I 7.0"" . i. .98.}! ......o...0.I...0..D..Q..X . .r . 3 .10 . . \ . . C. D... g Q . . 9 . .o 01. 5’3... . . . . . . . .. . . . . o k r. .....‘I ..I) .11.... u 21...! O 3!... I... . ...”...Jmulfluu. ...... ...-(dung. ..Q.b.|0_du. . ’.0x3f!..oJT. . . . .. . . . . . . . . {slut-“...? “Jon. 101‘. ...! 5.0... U... . 31......9... z. .5... 1.....8¢b.lu .Qvo.‘ r 0 o’; *.f:'¢.oo.,oooo: . . .0...‘ d. 0.2.}...3101‘1 . . 3..J..‘.‘. . . . . ... . . . . . . .. . . . t. .. . . . . 01.10.... .‘s B .‘3 . .0 .’{3..01 .....‘.0‘ 4.. ... o. I...f.0l.?.0.o6........800‘. ......ov roll... . . . . . .. . . . . .. .£.2‘0. . tor Yo’.$..uo . 0M0..’.0‘90.. .... v . a! ...-0“ a... .‘.;vfc‘!l.‘£ ......c "i...... . .. . I... .....Cq‘...‘. O. a .00.... ....IEI ... ode. . .44 . .... .‘0 ...... ._.3384.o‘....$§i8.‘33.2. ... ...!.IJ.....‘7CI. . . .. .. . .. I. .... 4.... 0" ..ofd....o....IJ._ . .. .10.... 2..-(.... . 5.0. ......l: ..FB....I.2..V.......J. .ll t....o....-..lt. . . ...91- ...OO! . .0. 3 ........:.l.0.o.l..nu"i C. 3.9. N... .3203 . . ...! .(I‘!!...u.o. . 3.33.!!! 3 0’0... I; 0’. I. . ... ......t. 0. “t...vlll.. ...-...!!! 8.1.3:. ....o‘n.o:- . .-. :7 01.19....91fp. .\ [3‘41 . 3.3.... .. .... ..v’q........¢..l.u.lloa.o:..oou .....‘1.... I... . .1 0.. :9). 01.0 .00.... 93‘? $3.13.... .1317... 0Q: 0315'. v. 0...... .123..." . .3“. l a I ......Q... 09...: ..I .....‘iivtuc. $5.01,). . . .0. .I .... 4 u ...-.1!‘ 0.530... . ........J0fi413....o.. .l ..l .32.. 3 .Ict'vt 0.0.7.... ..0 . . 1.0.3. o .0. 2.3.0.. .....1‘: .. ‘ .....11 ... 9.0 .3 .....o 0 00.1.0...‘ . ...) ......lcza. . .15.... ...... - 3......395 o... . 3o . . 4.. ......f. .’ ... .OH 2.41:... 0...! I4... 3.0... .... ..f . ....VIO" .‘of I09: . .. . oh...‘9t.’ J . . .. . .. o .. .0..... 0...::-’ .. . . .h . . . . . . . . . .. '3, I“. Q ..‘1 3.1.... .. . . . . . ... . . . .. .. ..00 SID-'11... .. . . . S...00.Oo.nl.¢ 5.0.. o. .0. 0.0 ....b. .... . ’3 ....vu... .. _ 2...... .99? .. J 7‘20." '0 ...- ... .o.r.. ...-.. 0 v 0 . 8.0.710; 1.3 «...-03-001.25.223. . . . .9 .c ..1. ...v’ 10.0 ..Q“ .1» . .'..E:fc. ......P.‘.o...0... 0.30.0. 4.0... . .0 A I ...... ...-5‘ Q 1119. ...... ”! o. 13...: 0.. o...” o’...9‘ . )fll! ... ... 001.1...u1). . 2v!.§...ii..$!'§0. 06“. t... . . “IMHO-”29‘ ...... .IO—u‘.‘.u': Ir’3‘ho" .‘ I .O 0‘ I .n“ ...:Lrom! i ......bxtl . .0 1’ ...... . '.0.’...OO 0‘. ... 0... a. .. ‘ .0. 3'1... . ." ...- ..pq‘. ..--r... ...-I... .... . .00... I... ....q!....,ssoo..’. .. 23.30....3. 20.09:, I ......‘0. . . I . . .1 .l . . t v . . I . . .- ...} . . . ... '3'....~7 . .. .....29. .IO ... 09.0. 99 . ... 4.... 60"“, J..‘hl"..i¥. 6.01,. . 390... ...}? .. . . . .... 0.90.5.0... L. .v’£C.:I . .. t. .. . . . . . . . . . . .....1..3.....0. .... . ... .. . . . .. . . . . . .1 ......'Q. .. ... . . . . . . . _ . . . .. ..Qtt: .. .. I .02.... . .. 2..-é... _ 0‘0... ..i .‘i... ...... ... . .... ...: . . . ... . U... a v 0 u n ‘1‘. .. . ... . kgtl..v . o .0 . . . z . . . .. .t -..-o .1 . ...... . . . . I .. . . . _ ... . ..— ..o.’ .1)... ZN"... . . . . . . .. . ...... .. .... £950.... . é . . ... . . . .. . . . . I.V’wr“.fl..uh1lu. t Id...a..nflp..o.\0...cul .1... 0 . . . . . .33-...u2... 3...... .. . . . . . . . . . .. .0 O'I’. ...0 0.. ... . ...of ...-0'.€ao..‘..... . . . .300... . ....0l . . . . . . -..! ‘0‘}. ...,1 3. I'- z . .... ...At. .0 ... Q . . ...-0 ..ol . ’0... I . ......h... : . : £1.24.-. . .. . ._ . . . __ . _ . . 1....-. .. ..u..u..z.z.... .n...n.................:......un..r.§2s§.. ‘60-...O.o C ..0 . 9|. .OQ‘.§.D..§‘.‘... . . . . . . . . . . . . 0.9... ...-0.... .0... 1'... 0......10.’H-A 0 ..."O ’snuoczg’ao ‘1 '10:... ....)3099 ..0 . . '00 D ...o‘:.l....'.0...v.|.o. .. . . . . . . . . . . . . . . . . . . .. . o. .o 9. r ......“C'l’o’. ......CoO. ..‘3... 3‘," O. 3; a ......OK‘II .0. v... ... a. . .. . . . .. . . .. y A ... X . . n, 0.! 9."‘3 .‘9‘ ...-‘1 ..d: 0.. ..'CI ..$ '2‘ . . . . . . . . . . .. . .. .. .. . . 90.30....3'9. 3.. a... .. . . .0 . :0 .0... 00.. $0.. ... . .. . . . . . .. 0. I .0.'. .g ....0 . 0 .....Dfoi ..-...o_.¢. .OI. . ..‘c' ...: . ...-.0304 b .‘I (0.0.1.3., . . . 0:. c . .5 o. I:"..:. .v... .. . . ..I. 3.1.. o .. . .... p A voill: . . . ...-.. 7.... ... . ... .1 .....T ... .9... Q . u 0.. I 0 ... a. o.‘. . . D. o. . up. 0 —. 00“; . . ... .... ... 3.1. 0.:. la ... o. .‘o .0. 1... O v. . .. . . .. . ...... 05.0... 1.. ‘. ..0 . ... .... 06.0.9...) No, . 0.. ......ti . . 7!! 4.00.... . . ...-cl... ... ‘..'.v~1’3 .J . 0 0. ... : 14...}... ...: . . o . .. ... '10... a... 1...... . .. 0... .....t....: . 0“..O.o . .. . . I... .... 0! . ...... 2.0.... . . . ..f . . . . . . .3" .3..- . .. . .... v 5.5.1. .....rl...0.....0. . : u ,5... ... . . .. . “$63.0... ... 00.... ... ...! .. .. 9. . . .. .056.Y.v”$l.3,~..0.3."v . ’93: . . 13.." 01‘. u .34.. .0 0.. 0. : . 11.5.6.3...3 .. .. . . . .25.... ... 3.. .. .. 91;... .. . . . ... 9‘30"“."3'39O! .9... ...... . ..Qc; OU'i .‘bur‘fi... ...sn , .... ..o ... . - .. ...... .I. . ... . 0 ... . .. ~32... Q..|£........:L.....¢. 1.225... . . . . . . . ..5. . 0.2352. l....9..cl..l,...:ru..6vt)! OviltO’Q .9. 0.. . I . ‘Ft‘:!. 0 .o «.02.... .... .301! ... t.. .O’ ‘06.. ...‘3... . 1. .. ...- .'.3......0 3.8 .... . . . . . .. .. .. 30.9.6... ‘3... -... ... A olo...l...!..fi. O.l".¢ .00 0.....‘!. _ ... . . u. o . 2 O... .....2‘. 0.0! 7...: .... ....0 v3... .... .. ... .32... ..v. .... .... 3 . .. . .I. .. l . .. _ .. . .. ._ . . . ....of. Of. .0.0.S.I.9...‘2 9....I0...931T”,I.l3.‘1«¥..’3 £00.. . A. . . .v 0 -.. r... . Q .01... .780... ... . 00. a) o . .02‘5 .. ...? . . . . ......l {2.2.3.1: .. ...... ...-...... ... o ......>.. . . . . . 2...... ....v...4.'..0 vol: ...". .... T... ..I 3.04.... . v . . .. | ‘I .. _. .... d .l. A ... .... 03.... .0 . 0 A . . .. .90. .Jic. . . . . . \..... ...... o v. .....9. l l . 33.4. . . ...... . . ...-... . . . 0a..- .... o... 5.0..v .. ‘o. . . .... ... .. o O l v . o 9 .I.¢. ....-. 1... 1:! 1.7.1.4 . . 6 510...! ll. .. .... 5...} ...; ... ... 0 .... . .Ov.t........ ...... . .. to... .. .. .. . I 0 lo. .... . O . . .. o 9 ..IO. 0.. .... .0. o .1. J . ... ......J. .35.... ... o .....93..... .OP‘FLQI 0.. .69 ..Oo. ....an .. A... . s. ...... . .. ..... . . .. c. .. . . u. a-.. o . A . . a . . 30 .... J . 0.. a . . 3.9.. ‘....0.cl.;... ...Oto . .. . O ... .. 0.... ...-I... 0.. .... .... ...... l3... :3... . . .08. ......» ‘0‘ . o .0 .00... ... .-. 40A'.J.1.A.. 90 ..’. . o... ...... . 0.9.... ... ~ 5 ..0 an..¢..,.£.. ......t.’ 01' ....OJ0 . . . .0...o... . . o 1.)... I 0.; o. ..-- u . ...-900.... 00.0) ...! .... .4. .. ...... o .0. ..v 2... . .03... o .10.. 5.6. in l9... 0 . '3 o . . . o .l' ... 0.. . . J ....A ... . I... 1..." u . ......O..- r. o I... . ... I .... .. .. .... .. .. O . .-§.I 4‘ .0 04.1 .0 .....II. ..V-‘.Ql¢.fi.9| 00. . ....o o‘ciol. .0. . . .. l. 9.... A.» .. v...v0..1 .94. V ... a . 0.. 0...... . l . . 0 0 . ‘9 . . . r. ‘15.. It . .94. . ....flocl . I . t,.'. ." Ir. ...0. ......I . o o . .... ..-. . . . n . ...... 1... . .... . . . .. ... 2150.00.10 .. .0. .. ... . I... a .|.. ...! 3 . a . . ....7 .... 4D... . ... . . . 0 . . . .. . ..- v- .I b.. 91...! .. ......o.....0-OO.O.. . o . .00. T .0.. .§ 0.... .u ... ...: - '00. o. .. . .. . . .. a”. .9..._.n. L... 0...... 1. . .6 0.1 ...o . ...... 0.. . 13.01 . 1.0! 0...... $90.... .0 ...l . 09‘. ...-.... o. . .1121... ...V . . -. -...‘vo 0 . .. _ .. .. . ... .. . 0.1. . .. . . ' ... V... . 2.1.... . . ...'.\.A.Y. . . ..I‘..OI~..-"ro........n.;~ 'Oo.‘0. ... .0... . A i... .... . .... U.‘...' . . . . . , .. . . . . . -... ... . .. ..- .....i... . . 0.. . . .. ... .....0. 6... Ti... .. V 3“. £10.. .10. . . J ..0 .....1... .. ...... it. . 3 .....X. .I..... (2..... .. . . . .. . . . 0". .P‘IOI .-v...... ..v.. . . . .41. 0.. no .oI...-0.v..4.. O. .05 .V. I- .- .. .. 0' . . 0......‘. 2‘. . ...: .100... .c . . . . .1'. _ L I o. .....«O. .....I‘ I. . o 0 o . 0.. Q . ..A u ‘0 . u .03». f... 3‘ .7..‘..o'-.U.Q ...... .6 . . O. .0 .u 0.. or... .o . .u . I. . .. 2' .‘ _ 01... ; co . ' of. ..‘IO. ....fi.....00¢9.........l . 0. ...-01 (2.3.4... .. . ... . . I. .. 0..-..1 44....- . » .... . . o . 0.... 0 .0 . ...». ...??? . . I.‘ . . . V...“Y.1.a.. ...l...... ....J 1......"- ...-.....930' I... .. ..o. I...... 3... .0 ... ._ . . .. I. ...l.‘ ... ....v. 4 0.. 0.1.51.0" .... 4" 110.31.. .0. 14.... ...9. O . 6|... 9. . v0. ‘. . ... . . . .3 .0... .u ...-17;... 354$ I .... ... o .. . . I. O... . ... . . .0 tl'. . 0 . 0...! .n .. '0 0.6 o. 433.. 5.0.5.... 5 A. .l‘. ...-I. .6 w T .. ...X a. .0 ... ... v..._ .. .. ......a... .0... I. . ... 79...... _ 0...-.. ......fi' .. ... .. ...0: 0 . ... . 0.3.0.... . 00.9--.. .9 l....oo ... n. .701...- . .X .0..'. .0 . ..I o 0. .. v 10031.... . o 60...... u . . . . . .9b.‘ ..0. 0' 0.. .0 0).. .... g. ...-3!. .. 2.5.30... .... 'o I... . $0.... .' . . . .01. ... . . .00. .. ... . . .. 6... I. .. . . . 1...; . .2 £01.. .0‘ .... .. .X. 0.. . l 0'. .0. ‘0 . .... -.. .- o...... .. ......C. .0. ~0. .00. ... .. ‘2‘...- . 6 - C It 31‘ ‘0 v.. I .. 1 .I ...9 .. . . ..l 0. . . ... .... ...... . I I C. 0.... .1.-Ddc.. ‘ .... co .‘ol'..o..a ...}... ... 50.. :3. .. .. I .0. a .9 d: ..v .1 3.. _. .. .0.Qo.. . . ..‘-J.i.‘€. '0‘: . . . . 90% . ... .0 0 .Ol.1¢:. t..- v. ......) .II. no. ..a............ . ! 50.... . _ . : .. 05 ..Il.’ ....100. ....D V: . . . ........ ..0. . ‘3-.J'..oo.i . . . .. . . . .O. . .. ..{A‘ .” ...... ... . .... . 1.0.10.6... ...-..0: . . . . . I. '1. .V .1. '1- . ....C......r.... . . . ...t. . ...... ... . . . .... ..V.!J.‘..... I). ltt.v. ... $0.0... ....C .. ..9? .9. .....o. . . ... . I: .l ...va O O. .. $3.... in ... -1-.. 0 . A3..- ...-9.. 0... .0). .. pt . - .. . . . . .. . . ..l ...: |O.....0.’.. .405‘.-. .... .. -9730... .... ...}... 0. .... .n a. ‘0‘»..- J . .. . .. . . . ... .. ... 0.94 o. 48' 0 .. ..A. . . . . ...! o . . ... . .. o . o _. . . . I. . ... ... . . . . .. .. .. . . 90341;. lt..§."‘io"o.§0‘ .10--.. .D ‘.c..v...' . .0. ...0 .... .....V‘ . .00.. .. . .. O... Q... 5.0.? NJ. 'I .9 Y‘ . . .o . 3'2: . It . .. c 01' 004. on. 0.3!.9‘.’ I... ttfio.9 . .. Pha‘..§.c.0'fld.¢§t’l’fit’lt . (04.0...“ .. 0.0. .9. .0. 3.13.7. l .. 9o... .I..0vvi...6.?o..r ... ......c.-.’ ...!!‘Q’OAK .391". on o I. ....0 . .... . .9...0..'...0u. I.OI..F!.IO_ ...... ..I._3....OOOSI..I‘\3.I ..lfib- 2!. . I -..‘0. .... . o . ...... . . ....90...‘ .00 . . . . 90.! ......II t.“. ‘O 6.0.. ‘0 OI. - '.'6O 9.! ..0 Hoots. . 0 v . I. .. .. . O T 19.... 2' 9 \\ . . . "¢..It.“ f t*l Cu ......ldoiili -‘ '1... .IV...‘ ..I . .' ... . _ . _. . . .... ’IO‘C‘.7- ......3 . 9‘ 5|. ll}..- . 0Q 06 10100.}! I- - .v to..' v.0 ‘ 0. On. I. . I s. . .. . . ...|¢ ...:l...’ n- .00‘ II ' . .0 ..-. 0' . .0... 00.0 006 '.’... A . o . .0 ......u .. .. .. . .....9. ..§.v.0”.. a b. l o 9.. I 1:; u... ’bblftl I'm 1 t '0 0" n r .v....n.3‘.'. .. . . .20... . . ... .I‘. . . .....0 ...-n! I! 6 ... . . . . . .O‘.0.. .... a. . ... .... ... ... . . I :1 0b...v '. .-.. .. '. 0. . . .0. .- .YQO . v0 .... you .6 ....c .... o. ...-.. II. I 1.... .c ..v. ... . 0.... ...O ...». ......otcb... .....‘l‘; ... . P. .39 v . .5‘ Q. 0.. 0. v . .. .-.. .0| 0. ... . . v ..Q. ..2........ol .. 0. ...... . . .. . 0’ D This is to certify that the thesis entitled INFLUENCE OF EVAPORATION AND MATRIX INTERFERENCES ON THE ASSOCIATION AND DISCRIMINATION OF IGNITABLE LIQUIDS USING CHEMOMETRIC PROCEDURES presented by Tiffany Paige Van De Mark has been accepted towards fulfillment of the requirements for the Master of degree in Criminal Justice Science [mm-i Major Professor’s Signature 21! (L AUGUST 2I/0_ Date MSU is an Affirmative Action/Equal Opportunity Employer LIBRARY Michigan State University u-n-o-I-o-n-I-I-'-a-I—I->-I-I-n- --.-o---—.---c—n-o-n—o-u-o-v-o-o-q-n-.---.--n-s-u---v-u—o-.-.-.‘.-.--c-.-- PLACE IN RETURN BOX to remove this checkout from your record. To AVOID FINES return on or before date due. MAY BE RECALLED with earlier due date if requested. DATE DUE DATE DUE DATE DUE 5/08 KlProj/AccaPres/CIRCIDateDue.indd INFLUENCE OF EVAPORATION AND MATRIX INTERFERENCES ON THE ASSOCIATION AND DISCRIMINATION OF IGNITABLE LIQUIDS USING CHEMOMETRIC PROCEDURES By Tiffany Paige Van De Mark A THESIS Submitted to Michigan State University In partial fulfillment of the requirements For the degree of MASTER OF SCIENCE Criminal Justice 2010 ABSTRACT INFLUENCE OF EVAPORATION AND MATRIX INTERFERENCES ON THE ASSOCIATION AND DISCRIMINATION OF IGNITABLE LIQUIDS USING CHEMOMETRIC PROCEDURES By Tiffany Paige Van De Mark In arson investigations, fire debris is collected, extracted, and analyzed by gas chromatography-mass spectrometry (GC-MS). The resulting chromatograms are visually assessed to identify an ignitable liquid residue (ILR). However, the ILR may differ in chemical composition from the original neat liquid due to evaporation and thermal degradation as well as interference effects from the burned debris. As a result, visual inspection of the chromatograrns is highly subjective and prone to misinterpretation. The goal of this research is to develop an objective method using chemometric procedures, such as principal components analysis (PCA), Pearson product moment correlation (PPMC) coefficients, and hierarchical cluster analysis (HCA), to identify an ignitable liquid in the fire debris despite evaporation, matrix interferences, and combustion. The first study investigated successful association of six ignitable liquids at four levels of evaporation to the corresponding neat liquid standard using PCA, PPMC coefficients, and HCA. In the second study, ignitable liquid mixtures containing gasoline and kerosene at two levels of evaporation were investigated to assess the ability to associate to the corresponding mixed liquid standard, despite the presence of evaporation, matrix interferences, and combustion. Using chemometric procedures, such as PCA, PPMC coefficients, and HCA, the mixed liquid samples were unable to be associated to the mixed liquid standard. This thesis is dedicated to my Granny, Virginia Vogel Van De Mark. I love you so much and miss you dearly. iii ACKNOWLEDGEMENTS During the past two years, many people have supported me in my journey of completing my master’s degree from Michigan State University. First, I would like to thank Dr. Ruth Smith for her assistance and more importantly, guidance throughout my research and thesis writing. Thank you for not giving up on me, even when I was frustrated. You have made me a more capable chemist and a better person through this process. I would also like to thank Dr. Victoria McGuffin for being one of my committee members and more importantly, an invaluable asset in my research and my thesis. I am extremely grateful to you for your support through this process. Thank you to Dr. Steve Dow for sitting on my committee as my Criminal Justice faculty member. Without the support of my friends both outside and inside the Forensic Science Program, I would not have finished my master’s degree. You all have been there for me every step of the way. Thanks to Kaitlin Prather, John McIlroy, Monica Bugeja, Melissa Bodnar, Ruth Udey, Christine Hay, Kari Anderson, Seth Hogg, Jamie Baemcopf, Patty Joiner, Beth Shattuck, Jacqueline Eufrausino, Kamilah Ziodeen, Nicole Pearl, Justine Hosein, Kathryn Walsh, Jaclyn Donlevie, and Richard Rarnmo for all of your love and support. Finally, I would like to thank my mother and my Sister. We have been through so much in the past two years, and I would not have been able to get through all of it without you both. Brittany, you have grown up so much, and I am proud that you are my sister. Mom, you have been such a great support through all of this, and I am so blessed to be able to call you my mother and my friend. I love you both so much. iv TABLE OF CONTENTS List of Tables viii List of Figures xii Chapter 1 — Introduction 1 1.1 Classification of Ignitable Liquids ................................................................................. 1 1.2 Extraction of Ignitable Liquid Residues from Fire Debris ............................................ 3 1.3 Analysis of Ignitable Liquid Residues ........................................................................... 6 1.4 Problems in Identifying Ignitable Liquid Residues in Fire Debris ................................ 8 1.5 Literature Review ........................................................................................................... 9 1.5.1 Procedures to Reduce Non-Chemical Sources of Variance in Chromatographic Data .............................................................................................................................................. 9 1.5.2 Evaporation of Ignitable Liquids .............................................................................. 13 1.5.3 Matrix Interferences .................................................................................................. 16 1.5.4 Statistical and Chemometric Analysis of Ignitable Liquids ..................................... 21 1.6 Research Objectives ..................................................................................................... 24 1.7 References .................................................................................................................... 27 Chapter 2 — Instrumental, Statistical, and Chemometric Techniques 30 2.1 Gas Chromatography-Mass Spectrometry ................................................................... 30 2.2 Data Pretreatment Procedures ...................................................................................... 41 2.2.1 Savitzky-Golay Smooth ............................................................................................ 41 2.2.2 Retention Time Alignment ....................................................................................... 43 2.2.2.1 Target Selection ..................................................................................................... 43 2.2.2.2 Peak Matching Algorithm ...................................................................................... 44 2.2.2.3 Correlation Optimized Warping Algorithm ........................................................... 45 2.2.3 Normalization ........................................................................................................... 47 2.3 Principal Components Analysis ................................................................................... 48 2.4 Pearson Product Moment Correlation Coefficients ..................................................... 51 2.5 Hierarchical Cluster Analysis ...................................................................................... 52 2.6 References .................................................................................................................... 56 Chapter 3 — Association of Evaporated Ignitable Liquids to Their Neat Counterparts Using Chemometric Procedures <7 3.1 Introduction .................................................................................................................. 57 3.2 Materials and Methods ................................................................................................. 58 3.2.1 Sample Collection ..................................................................................................... 58 3.2.2 Sample Preparation ................................................................................................... 58 3.2.3 GC-MS Analysis ....................................................................................................... 60 3.2.4 Data Pretreatment ...................................................................................................... 61 3.2.5 Data Analysis ............................................................................................................ 63 3.3 Results and Discussion ................................................................................................ 65 3.3.1 Optimization of Retention Time Alignment ............................................................. 65 3.3.2 Normalization ........................................................................................................... 71 3.3.3 Association and Discrimination of the Neat Liquids using PCA ............................. 71 V 3.3.4 Association of Evaporated Liquids to Neat Liquids using PCA ............................... 80 3.3.5Association and Discrimination of Evaporated Liquids to Neat Liquids using PPMC coefficients ......................................................................................................................... 84 3.3.6 Association of Evaporated Liquids to Neat Liquids using HCA .............................. 89 3.4 Conclusions .................................................................................................................. 92 3.5 References .................................................................................................................... 94 Chapter 4 — Effect of Matrix Interferences, Evaporation, and Combustion on the Identification of Mixed Ignitable Liquids in Fire Debris using Chemometric Procedures 95 4.1 Introduction .................................................................................................................. 95 4.2 Materials and Methods ................................................................................................. 96 4.2.1 Sample Collection ..................................................................................................... 96 4.2.2 Mixed Ignitable Liquids ............................................................................................ 96 4.2.3 Matrix Interferences .................................................................................................. 98 4.2.4 GC-MS Analysis ....................................................................................................... 99 4.2.5 Data Pretreatment ...................................................................................................... 99 4.2.6 Data Analysis .......................................................................................................... 101 4.3 Results and Discussion .............................................................................................. 102 4.3.1 Optimization of Retention Time Alignment and Normalization ............................ 102 4.3.2 Matrix Interferences ................................................................................................ 106 4.3.3 Association and Discrimination of the Mixed Liquids in the Presence of Matrix Interferences and Evaporation ......................................................................................... 110 4.3.4 Association and Discrimination of the Simulated ILRS in the Presence of Matrix Interferences, Evaporation, and Combustion ................................................................... 118 4.4 Conclusions ................................................................................................................ 125 4.5 References .................................................................................................................. 127 Chapter 5 — Conclusions and Future Work 128 5.1 Association of Evaporated Ignitable Liquids to Their Neat Counterparts Using Chemometric Procedures ................................................................................................. 128 5.2 Effect of Matrix Interferences on the Identification of Mixed Ignitable Liquids in Fire Debris using Chemometric Procedures ............................................................................ 131 5.3 Future Work ............................................................................................................... 133 Appendix A - Total Ion Chromatograms of Neat Ignitable Liquids 135 Appendix B — Total Ion Chromatograms of Evaporated Ignitable Liquids ............ 139 Appendix C - Pearson Product Moment Correlation Coefficients for Total Ion Chromatograms of Neat and Evaporated Liquids by Ignitable Liquid Class ......... 152 Appendix D - Pearson Product Moment Correlation Coefficients for Total Ion Chromatograms of Neat and Evaporated Liquids by Evaporation Level ................ 159 Appendix E - Total Ion Chromatograms of Mixed Ignitable Liquid Standards ....165 vi Appendix F — Pearson Product Moment Correlation Coefficients for Total Ion Chromatograms of Burned Carpet, Mixed Liquid Standards, and Burned then Spiked Samples 169 Appendix G — Pearson Product Moment Correlation Coefficients for Total Ion Chromatograms of Burned Carpet, Mixed Liquid Standards, and Simulated Ignitable Liquid Residue Samples 175 vii LIST OF TABLES Table 1.1 ASTM ignitable liquid classes and major components within each class ........... 2 Table 1.2 Examples of ignitable liquids in each ASTM class ............................................. 4 Table 3.1 Ignitable liquids investigated ............................................................................. 59 Table 3.2 User defined parameters investigated for peak matching algorithm and COW algorithm ............................................................................................................................ 62 Table 3.3: PPMC coefficients for all replicates (n=15) for optimal peak matching alignment and Optimal COW alignment ............................................................................ 70 Table 3.4: Mean PCA score for replicates (n=3) of each liquid class based on the TIC ...74 Table 3.5: PPMC coefficients for replicates (n=3) and range of coefficients for each liquid class based on the TIC ............................................................................................. 85 Table 3.6: PPMC coefficients (n=9) between neat and evaporated gasoline and neat and evaporated lacquer thinner ................................................................................................. 88 Table 4.1 Mixed Ignitable Liquids ..................................................................................... 97 Table 4.2 User defined parameters investigated for peak matching algorithm and COW algorithm .......................................................................................................................... 100 Table 4.3: PPMC coefficients for replicates (n=3) for the mixed liquids and burned carpet based on the TIC .............................................................................................................. 112 Table 4.4: PPMC coefficients between burned then spiked samples and corresponding mixed liquid standard (n=9) and burned then spiked mixed liquid and burned carpet (n=9) based on the TIC .............................................................................................................. 117 Table 4.5: PPMC coefficients between simulated ILR sample and corresponding mixed liquid standard (n=9) and the simulated ILR sample and burned carpet (n=9) based on the TIC ................................................................................................................................... 123 Table C.l: PPMC Coefficients for Replicates (A, B, C) of Lamp Fuel at Neat (N), 5% Evaporated (5), 10% Evaporated (10), 20% Evaporated (20), 50% Evaporated (50) ..... 153 Table C.2: PPMC Coefficients for Replicates (A, B, C) of Kerosene at Neat (N), 5% Evaporated (5), 10% Evaporated (10), 20% Evaporated (20), 50% Evaporated (50 ...... 154 viii Table C.3: PPMC Coefficients for Replicates (A, B, C) of Marine Fuel Stabilizer at Neat (N), 5% Evaporated (5), 10% Evaporated (10), 20% Evaporated (20), 50% Evaporated (50) .................................................................................................................................. 155 Table C.4: PPMC Coefficients for Replicates (A, B, C) of Paint Thinner at Neat (N), 5% Evaporated (5), 10% Evaporated (10), 20% Evaporated (20), 50% Evaporated (50) ..... 156 Table C.5: PPMC Coefficients for Replicates (A, B, C) of Lacquer Thinner at Neat (N), 5% Evaporated (5), 10% Evaporated (10), 20% Evaporated (20), 50% Evaporated (50)157 Table C.6: PPMC Coefficients for Replicates (A, B, C) of Gasoline at Neat (N), 5% Evaporated (5), 10% Evaporated (10), 20% Evaporated (20), 50% Evaporated (50) ..... 158 Table D1: PPMC Coefficients for Replicates (A, B, C) of Neat Ignitable Liquids, Gasoline (Gas), Kerosene (Kero), Lacquer Thinner (LT), Lamp Fuel (LF), Marine Fuel Stabilizer (MFS), Paint Thinner (PT) .............................................................................. 160 Table D2: PPMC Coefficients for Replicates (A, B, C) of 5% Evaporated Ignitable Liquids, Gasoline (Gas), Kerosene (Kero), Lacquer Thinner (LT), Lamp Fuel (LF), Marine Fuel Stabilizer (MF S), Paint Thinner (PT) .......................................................... 161 Table D3: PPMC Coefficients for Replicates (A, B, C) of 10% Evaporated Ignitable Liquids, Gasoline (Gas), Kerosene (Kero), Lacquer Thinner (LT), Lamp Fuel (LF), Marine Fuel Stabilizer (MFS), Paint Thinner (PT) .......................................................... 162 Table D4: PPMC Coefficients for Replicates (A, B, C) of 20% Evaporated Ignitable Liquids, Gasoline (Gas), Kerosene (Kero), Lacquer Thinner (LT), Lamp Fuel (LF), Marine Fuel Stabilizer (MFS), Paint Thinner (PT) .......................................................... 163 Table D5: PPMC Coefficients for Replicates (A, B, C) of 50% Evaporated Ignitable Liquids, Gasoline (Gas), Kerosene (Kero), Lacquer Thinner (LT), Lamp Fuel (LF), Marine Fuel Stabilizer (MF S), Paint Thinner (PT) .......................................................... 164 Table F.1: PPMC Coefficients for Replicates (A, B, C) of Burned Carpet (BC) and Mixed Liquid Standards, Neat gasoline: Neat Kerosene (NGzNK), Neat Gasoline: 10% Evaporated Kerosene (N G:10K), Neat Gasoline: 50% Evaporated Kerosene (NG:50K), 10% Evaporated Gasoline: Neat Kerosene (10G:NK). 50% Evaporated Gasoline: Neat Kerosene (50G:NK) ......................................................................................................... 170 Table F.2: PPMC Coefficients for Replicates (A, B, C) of Burned then Spiked Samples, Neat gasoline: Neat Kerosene Sample (BNGzNK), Neat Gasoline: 10% Evaporated Kerosene Sample (BNG210K), Neat Gasoline: 50% Evaporated Kerosene Sample (BNG:50K), 10% Evaporated Gasoline: Neat Kerosene Sample (BlOGzNK). 50% Evaporated Gasoline: Neat Kerosene Sample (B50G:NK) ............................................. 171 ix Table F.3: PPMC Coefficients for Replicates (A, B, C) of Burned Carpet (BC) and Burned then Spiked Samples, Neat gasoline: Neat Kerosene Sample (BNGzNK), Neat Gasoline: 10% Evaporated Kerosene Sample (BNG:10K), Neat Gasoline: 50% Evaporated Kerosene Sample (BNG:50K), 10% Evaporated Gasoline: Neat Kerosene Sample (B10G:NK). 50% Evaporated Gasoline: Neat Kerosene Sample (B50G:NK)...172 Table F.4: PPMC Coefficients for Replicates (A, B, C) of Neat Gasoline: 10% Evaporated Kerosene Mixture (NG:10K) and Corresponding Burned then Spiked Samples (BNG: 10K) ........................................................................................................ 173 Table F.5: PPMC Coefficients for Replicates (A, B, C) of Neat Gasoline: 50% Evaporated Kerosene Mixture (NG:50K) and Corresponding Burned then Spiked Samples (BNG:50K) ........................................................................................................ 173 Table F.6: PPMC Coefficients for Replicates (A, B, C) of 10% Evaporated Gasoline: Neat Kerosene Mixture (10G2NK) and Corresponding Burned then Spiked Samples (B10G:NK) .................................................................................................................... 174 Table F. 7: PPMC Coefficients for Replicates (A, B, C) of 50% Evaporated Gasoline: Neat Kerosene Mixture (50G: NK) and Corresponding Burned then Spiked Samples (BSOG: NK) ...................................................................................................................... 174 Table G.1: PPMC Coefficients for Replicates (A, B, C) of Simulated ILR Samples, Neat gasoline: Neat Kerosene ILR Sample (SNG:NK), Neat Gasoline: 10% Evaporated Kerosene ILR Sample (SNG:10K), Neat Gasoline: 50% Evaporated Kerosene ILR Sample (SNG:50K), 10% Evaporated Gasoline: Neat Kerosene ILR Sample (SlOGzNK). 50% Evaporated Gasoline: Neat Kerosene ILR Sample (S50G:NK) .............................. 176 Table G.2: PPMC Coefficients for Replicates (A, B, C) of Burned Carpet (BC) and Simulated ILR Samples, Neat gasoline: Neat Kerosene ILR Sample (SNG:NK), Neat Gasoline: 10% Evaporated Kerosene ILR Sample (SNG:10K), Neat Gasoline: 50% Evaporated Kerosene ILR Sample (SNG:50K), 10% Evaporated Gasoline: Neat Kerosene ILR Sample (SlOGzNK). 50% Evaporated Gasoline: Neat Kerosene ILR Sample (S50G:NK) ....................................................................................................................... 177 Table G.3: PPMC Coefficients for Replicates (A, B, C) of Neat Gasoline: 10% Evaporated Kerosene Mixture (NG:10K) and Corresponding Simulated ILR Samples (SNG:10K) ....................................................................................................................... 178 Table 6.4: PPMC Coefficients for Replicates (A, B, C) of Neat Gasoline: 50% Evaporated Kerosene Mixture (NG:50K) and Corresponding Simulated ILR Samples (SNG:50K) ....................................................................................................................... 178 Table G.5: PPMC Coefficients for Replicates (A, B, C) of 10% Evaporated Gasoline: Neat Kerosene Mixture (10G:NK) and Corresponding Simulated ILR Samples (SlOGzNK) ....................................................................................................................... 179 Table G.6: PPMC Coefficients for Replicates (A, B, C) of 50% Evaporated Gasoline: Neat Kerosene Mixture (50G:NK) and Corresponding Simulated ILR Samples (S50G:NK) ....................................................................................................................... 179 xi LIST OF FIGURES Figure 2.1: Diagram of GC inlet ........................................................................................ 32 Figure 2.2: Diagram of GC column in temperature controlled oven ................................. 34 Figure 2.3: Diagram of electron ionization of compounds in mass spectrometer ............. 36 Figure 2.4: Diagram of quadrupole mass analyzer ............................................................ 38 Figure 2.5: Diagram of continuous-dynode electron multiplier ........................................ 40 Figure 2.6: A) Diagram depicting the single linkage method used in HCA. B) Diagram depicting the complete linkage method used in HCA ....................................................... 54 Figure 3.1: A) Poorly aligned 2,6-dimethylundecane peak in the TIC of neat and evaporated marine fuel stabilizer using a window size of seven. B) Well aligned 2,6- dimethylundecane peak in the TIC of neat and evaporated marine fuel stabilizer using a window size of three .......................................................................................................... 66 Figure 3.2: A) Poorly aligned 2,2,3,5-tetramethylheptane peak in the TIC of neat and evaporated paint thinner with a warp size of six and a segment size of 45. B) Well aligned 2,2,3,5-tetramethylheptane peak in the TIC of neat and evaporated paint thinner with a warp size of three and a segment size of 65 ....................................................................... 68 Figure 3.3: A) Poorly aligned 2,2,3,5-tetramethylheptane peak in the TIC of neat and evaporated paint thinner with a window size of three using the peak matching algorithm. B) Well aligned 2,2,3,5-tetramethylheptane peak in the TIC of neat and evaporated paint thinner with a warp size of three and a segment size of 65 using the COW algorithm ..... 69 Figure 3.4: A) Unnormalized pentylcyclohexane peak in the TIC of neat and evaporated marine fuel stabilizer. B) Maximum peak normalized pentylcyclohexane peak in the TIC of neat and evaporated marine fuel stabilizer. C) Total area normalization of replicates of the pentylcyclohexane peak in the TIC of neat and evaporated marine fuel stabilizer. Neat (—), 5% Evaporated (— u), 10% Evaporated (— a n), 20% Evaporated (— — ), 50% Evaporated (I I I) ................................................................................................................ 72 Figure 3.5: Scores plot of PCI vs. PC2 based on the TIC for the six neat ignitable liquids. Liquids were indicated by symbol. Lamp Fuel (I), Kerosene (0), Marine Fuel Stabilizer (0), Paint Thinner (O), Gasoline (A), and Lacquer Thinner (*) ...................................... 73 Figure 3.6: Loadings plots of (A) PC] and (B) PC2 based on the TIC for six neat liquids. Major components are labeled: 1) toluene, 2) ethylbenzene, 3) o-xylene, 4) p-xylene, 5)]- ethyl-3—methylbenzene, 6) 2,2,3,5-tetramethylheptane, 7) 5-ethyl-2,2,3—tn'methyheptane, 8) 3,6-dimethylundecane, 9) 2,6-dimethylundecane .......................................................... 77 xii Figure 3.7: A) TIC of lamp fuel. B) Mean-centered chromatogram of lamp fuel. C) Mean-centered chromatogram multiplied by the eigenvector for PCI .............................. 79 Figure 3.8: Scores plot of PC] vs. PC2 based on the TIC for six ignitable liquids with projections of evaporated liquids. Liquids are indicated by symbol. Lamp Fuel (I), Kerosene (0), Marine Fuel Stabilizer (0), Paint Thinner (O), Gasoline (A), and Lacquer Thinner (t). Each fill indicates a different level of evaporation. Neat (Filled e.g. I), 5% evaporated (Half Filled e.g. B), 10% evaporated (Cross e.g. E), 20% evaporated (Line e.g. B), 50% evaporated (No Fill e.g.El) ........................................................................... 82 Figure 3.9: HCA Dendrogram of the scores of the neat liquids. Dashed line indicates similarity of 0.959 .............................................................................................................. 90 Figure 3.10: HCA Dendrogram of the scores of the neat and evaporated liquids. An asterisk indicates 50% evaporated kerosene. Dashed line indicates similarity of 0.836 ...91 Figure 4.1: A) Unaligned toluene peak in the TIC of mixed liquids. B) Well aligned toluene peak in the TIC of mixed liquids with a warp size of two pts and a segment size of 75 pts ............................................................................................................................ 104 Figure 4.2: A) Unnormalized C14 normal alkane peak. B) Normalized C14 normal alkane peak. Neat Gasoline: Neat Kerosene Mixture (—), Neat Gasoline: 10% Evaporated Kerosene Mixture (- — ), Neat Gasoline: 50% Evaporated Kerosene Mixture (- u u )...105 Figure 4.3: A) TIC of burned carpet. B) TIC of 10% evaporated gasoline: neat kerosene mixture. Major components are labeled: 1) toluene, 2) 2,4-dimethyl-l-heptene, 3)styrene, 4) benzaldehyde, 5) acetophenone, 6) 1,3-diphenylpropane ........................................... 107 Figure 4.4: A) TIC of burned carpet spiked with 10% evaporated gasoline: neat kerosene mixture. B) TIC of 10% evaporated gasoline: neat kerosene simulated ILR sample. Major components are labeled: 1) toluene, 2) 2,4-dimethyl-1-heptene, 3) styrene, 4) benzaldehyde, 5) 1,3-diphenylpropane, 6) acetophenone ................................................ 108 Figure 4.5: Scores plot of PCl vs. PC2 based on the TIC for burned carpet and mixed liquid standards. Liquids were indicated by symbol. Burned Carpet (I), Neat Gasoline: Neat Kerosene Mixture(0), Neat Gasoline: 10% Evaporated Kerosene Mixture (at), Neat Gasoline: 50% Evaporated Kerosene Mixture (A), 10% Evaporated Gasoline: Neat Kerosene Mixture (O), and 50% Evaporated Gasoline: Neat Kerosene (O) ................... 111 Figure 4.6: Loadings plots of (A) PC] and (B) PC2 based on the TIC for burned carpet ad mixed liquid standards. Major components were labeled: 1) toluene, 2) 2,4-dimethyl-l- heptene 3) ethylbenzene, 4) o-xylene, 5) p-xylene, 6) styrene, 7) a-methylstyrene, 8) acetophenone, 9) 1,3-diphenylpropane ............................................................................ 113 xiii Figure 4.7: Figure 4.7: Scores plot of PCI vs. PC2 based on the TIC for the burned carpet and mixed liquid standards. Liquids were indicated by symbol. Burned Carpet (I), Neat Gasoline: Neat Kerosene Mixture (I), Neat Gasoline: 10% Evaporated Kerosene Mixture (t), Neat Gasoline: 50% Evaporated Kerosene Mixture (A), 10% Evaporated Gasoline: Neat Kerosene Mixture (0), and 50% Evaporated Gasoline: Neat Kerosene (O). The half fill indicates burned then spiked samples ........................................................................ 115 Figure 4.8: HCA Dendrogram of the scores of the data set containing the replicates (A, B, C) of the burned then spiked samples. The N is the neat part of the sample, and kero represents kerosene. The italicized labels indicate the burned then spiked samples ....... 119 Figure 4.9: Scores plot of PCI vs. PC2 based on the TIC for the burned carpet and the mixed liquid standards. Liquids were indicated by symbol. Burned Carpet (I), Neat Gasoline: Neat Kerosene Mixture(0), Neat Gasoline: 10% Evaporated Kerosene Mixture (at), Neat Gasoline: 50% Evaporated Kerosene Mixture (A), 10% Evaporated Gasoline: Neat Kerosene Mixture (O), and 50% Evaporated Gasoline: Neat Kerosene (O). No fill indicates simulated ILR samples ..................................................................................... 120 Figure 4.10: HCA Dendrogram of the scores of the data set containing the replicates (A, B, C) of the simulated ILR samples. The N is the neat part of the sample, and kero represents kerosene. The italicized labels indicate the simulated ILR samples .............. 124 Figure A. l: Chromatogram of lamp fuel with major components labeled ...................... 136 Figure A.2: Chromatogram of kerosene with major components labeled ....................... 136 Figure A.3: Chromatogram of marine fuel stabilizer with major components labeled 137 Figure A.4: Chromatogram of paint thinner with major components labeled ................. 137 Figure A.5: Chromatogram of gasoline with major components labeled ........................ 138 Figure A.6: Chromatogram of lacquer thinner with major component labeled ............... 138 Figure B]: Chromatogram of 5% evaporated lamp fuel with major components labeled .......................................................................................................................................... 140 Figure B.2: Chromatogram of 10% evaporated lamp fuel with major components labeled. .......................................................................................................................................... 140 Figure 3.3: Chromatogram of 20% evaporated lamp fuel with major components labeled. .......................................................................................................................................... 141 Figure 8.4: Chromatogram of 50% evaporated lamp fuel with major components labeled . .......................................................................................................................................... 141 xiv Figure 3.5: Chromatogram of 5% evaporated kerosene with major components labeled .......................................................................................................................................... 142 Figure 8.6: Chromatogram of 10% evaporated kerosene with major components labeled .. .......................................................................................................................................... 142 Figure 3.7: Chromatogram of 20% evaporated kerosene with major components labeled .. .......................................................................................................................................... 143 Figure 8.8: Chromatogram of 50% evaporated kerosene with major components labeled .. .......................................................................................................................................... 143 Figure B.9: Chromatogram of 5% evaporated marine fuel stabilizer with major components labeled .......................................................................................................... 144 Figure B.10: Chromatogram of 10% evaporated marine fuel stabilizer with major components labeled .......................................................................................................... 144 Figure B.11: Chromatogram of 20% evaporated marine fuel stabilizer with major components labeled .......................................................................................................... 145 Figure 3.12: Chromatogram of 50% evaporated marine fuel stabilizer with major components labeled .......................................................................................................... 145 Figure B. 1 3: Chromatogram of 5% evaporated paint thinner with major components labeled .............................................................................................................................. 146 Figure B.14: Chromatogram of 10% evaporated paint thinner with major components labeled .............................................................................................................................. 146 Figure 3.15: Chromatogram of 20% evaporated paint thinner with major components labeled .............................................................................................................................. 147 Figure B.16: Chromatogram of 50% evaporated paint thinner with major components labeled .............................................................................................................................. 147 Figure B.17: Chromatogram of 5% evaporated gasoline with major components labeled .......................................................................................................................................... 148 Figure B. l 8: Chromatogram of 10% evaporated gasoline with major components labeled. .......................................................................................................................................... 148 Figure B.19: Chromatogram of 20% evaporated gasoline with major components labeled. .......................................................................................................................................... 149 XV Figure 3.20: Chromatogram of 50% evaporated gasoline with major components labeled. .......................................................................................................................................... 149 Figure B.21: Chromatogram of 5% evaporated lacquer thinner with major component labeled .............................................................................................................................. 150 Figure B.22: Chromatogram of 10% evaporated lacquer thinner with major component labeled .............................................................................................................................. 150 Figure B.23: Chromatogram of 20% evaporated lacquer thinner with major component labeled .............................................................................................................................. 151 Figure 3.24: Chromatogram of 50% evaporated lacquer thinner with major component labeled .............................................................................................................................. 151 Figure E.1: Chromatogram of neat gasoline: neat kerosene mixture with major components labeled .......................................................................................................... 166 Figure E.2: Chromatogram of neat gasoline: 10% evaporated kerosene mixture with major components labeled ............................................................................................... 166 Figure E.3: Chromatogram of neat gasoline: 50% evaporated kerosene mixture with major components labeled ............................................................................................... 167 Figure E.4: Chromatogram of 10% evaporated gasoline: neat kerosene mixture with major components labeled ............................................................................................... 167 Figure E.5: Chromatogram of 50% evaporated gasoline: neat kerosene mixture with major components labeled ............................................................................................... 168 xvi CHAPTER 1 Introduction According to the United States Fire Administration (USAF), the estimated 30,500 intentionally set structure fires in 2008 killed 315 civilians and caused an estimated $866 million in property damage [1]. The criminal act of setting a fire with malicious intent to destroy the property of another is considered arson. Most commonly at an arson scene, ignitable liquids are used as accelerants in order to rapidly ignite and spread a fire in a given area. The presence of an ignitable liquid would indicate that the fire was intentionally ignited in order to cause extensive damage. To confirm the presence of an ignitable liquid in fire debris collected from a suspected arson scene, a fire debris analyst must identify the ignitable liquid through visual comparison to a known ignitable liquid. 1.1 Classification of Ignitable Liquids As defined by the American Society of Testing and Materials (ASTM), there are eight classes of ignitable liquids which, are classified based on the types of hydrocarbons present [2]. The major classes of ignitable liquids along with the chemical compounds that are present in each ignitable liquid are shown in Table 1.1. 82:58 82:58 m 8a 08 88.8 88: 89 882858 88 .a a Beszoeom a: .2 8088: flag—om Babe—n85 $38? 338880 38 88.36 685 .oeomobM E8830 o8om 85o 8338: 88:3me 83—8805 88.33 3528 $0855 0858.3 3328 383.68 18883 oEom 3mg engage—oxo 258: 32 £528 538 mm m .5 88.58am 8% 83:83.. macaw—o5 365888 o8om 08.8% ca 0885: 88 8880 a £08 0:830 888 .5an we :25 mung—om $0285 0882-2 8528 $838 8869:. we: 8 2 88 8883 3.5868 388.0 338 am 339-0523 o8om v m .8 2 20.40.? own—8 2: 8 £3 08330 258.5 been :88: .8: EEO .830 Some 853» 388888 8.88 we.» memmflo Eng “.3889 2Hm< : 033. Within each ignitable liquid class except for gasoline, three subclasses are used to separate the ignitable liquids based on boiling point, which relates to the number of carbons present in each hydrocarbon. The three subclasses are light, medium, and heavy. The light class has low boiling point compounds from C4-C9. The medium class contains compounds from C3-C13, which have boiling points higher than those of the light class. The heavy class has the compounds C9-C20+, which have higher boiling points than those of the medium class. If an ignitable liquid contains a wider range of hydrocarbons than a subclass, it can be classified into two subclasses, such as “light to medium” or “medium to heavy”. Shown in Table 1.2 are examples of ignitable liquids categorized into the corresponding classes and subclasses. 1.2 Extraction of Ignitable Liquid Residues from Fire Debris When fire debris is collected from a suspected arson scene, the crime scene investigator collects the fire debris, which contains multiple matrices, into a paint can. The fire debris is then submitted to a forensic laboratory for analysis. Upon receiving the evidence, the fire debris analyst opens the fire debris evidence container and writes a description of the evidence received. Then, the fire debris analyst decides on an extraction procedure in order to extract any ignitable liquid residue (ILR) from the fire debris. Three extraction procedures recommended by ASTM are solvent extraction, active headspace extraction, and passive headspace extraction [3-5]. 8820 8030.8 05 0:: E 80 00 85 88.0 58850 0088088 < 0088:0082 8000 588.50 888.0 850 5;» 808.— a 88.0 588 08880 00800995 88>.0m . . 00800995 8088 8085 8085 8085 8808 88000 88:85 88—0805 08 8885 . 808.8 308005 08585 8085 8 2 8085 82 080808 8085 880:0 00885 0805: 82 0088 0 880 0 008.. 0 8088 808005 0 50 2 a A: 2 0 A— u 2 888 8008.5 0850.8500— 888 0 88C 0 88.. 0 08858 .8088 0 0C 08 - 0 50 2 0 A: Z a A: 2 880—88802 0. :0 5 0= Detector Figure 2.2: Diagram of GC column in temperature controlled oven. 34 Along with boiling point, the interaction with the stationary phase of the analytes will affect separation, which is based on the polarity. The stationary phase can be nonpolar and made of a thin liquid layer of polymethysiloxanes. The thin layer coats the inside of the GC column and is no more than 0.25 micrometers thick [1]. When the compounds pass through the column, the analytes interact with the stationary phase. Furthermore, the nonpolar analytes have more interaction with the stationary phase than the polar analytes. So, the nonpolar analytes are retained on the GC column longer than those that are polar, thus causing separation. Once they have been separated in the GC column, the analytes pass from atmospheric pressure through a heated transfer line (300°C) into a vacuum. This is possible due to the low flow rate and to the GC column fitting directly into the detector. Of the many detectors available, the most common detector in a forensic laboratory is the mass spectrometer. The mass spectrometer contains three parts: the ionization source, the mass analyzer, and the detector. In the ionization source, the analytes are ionized and fragmented. Then, the fragments are separated by mass-to- charge (m/z) ratio in the mass analyzer, and are finally detected by the detector. The entire mass spectrometer is maintained under vacuum in order for the ions to remain stable. The mass spectrometers in forensic laboratories usually contain an electron ionization source, a quadrupole mass analyzer, and a continuous dynode electron multiplier detector. Of the many ionization sources available, electron ionization is the most common ionization used in forensic laboratories. Upon entering the mass spectrometer, analytes are ionized through a stream of electrons that are emitted from a filament (Figure 2.3). 35 Filament C01 0::5:: .5:::EEEE Mass inOven O o oo Analyzer l .. I Anode Figure 2.3: Diagram of electron ionization of compounds in mass spectrometer. 36 The electrons emitted from the filament typically have a voltage of 70 eV and are focused through the electron slit and travel towards the anode. A voltage of 70 eV is advantageous since the maximum energy from the electron is transferred to the analytes. As the stream of electrons is passed perpendicularly to the analytes, interactions occur between them. The primary interaction is where the analyte loses an electron and becomes positively charged, thus forming a molecular ion. Other positively charged fragments will also be formed from the molecular ion interacting with other ions or other electrons within the electron stream [1]. Once the analyte has been ionized and fi'agmented, the molecular ion and corresponding fragments are focused through electrostatic lenses and accelerated into the mass analyzer (Figure 2.4). For most forensic analyses and this research, a quadrupole mass analyzer is used. As inferred by its name, the quadrupole mass analyzer has four poles, or rods, that form a diamond shape inside the mass analyzer. The pairs of rods are orthogonal to each other and are oppositely charged with one pair containing a positive charge and the other pair containing a negative charge. The charges on the rods continually alternate. For each set of rods, the charge is applied from a direct current (DC) power source. Along with DC, radio-frequency (RF) voltages are also applied to the rods. As the ions enter and pass through the quadrupole, the DC and RF voltages are increased, while the ratio between them remains constant. As the voltages are increased, the ions oscillate in between the rods because of the alternating positive and negative charges [1]. Due to the increase in RF and DC voltages, some ions will develop unstable trajectories while passing through the quadrupole, impact the rods, and be neutralized. Other ions with stable trajectories 37 Ion with unstable "away Detector I" .27 Ion with stable trajectory Figure 2.4: Diagram of quadrupole mass analyzer. 38 will successfully oscillate between the pairs of rods and pass through the quadrupole to the detector. Each RF and DC voltage applied to the rods corresponds to an m/z ratio, thus increasing the voltages across the rods while keeping the ratio between them constant allows for a full mass scan (e. g. m/z 50-550). The detector within the mass spectrometer detects the ions that have passed through the mass analyzer. Of the detectors that are available, the continuous-dynode electron multiplier is one of the more commonly used detectors (Figure 2.5). It is made of glass that contains a high amount of lead which allows it to maintain an electrical potential of 1.8 to 2 kV [1]. The electron multiplier is shaped in the form of a curved cylinder that has a large opening on one side and narrows to a small enclosure on the other side. When the ions enter the detector through the larger opening, they are attracted to the negatively charged sides of the multiplier. Upon impact, the ions eject secondary electrons which are attracted to higher voltages present further into the multiplier. As the number of impacts increase, the number of secondary electrons increases exponentially [1]. After multiple impacts with the sides of the transducer, the secondary electrons produced are detected at the closed end of the cylinder. The output of GC-MS analysis is the total ion chromatogram (TIC), which is the sum of all of the ions detected in the mass spectrometer. The TIC is a plot of retention time on the x-axis and abundance of the analytes on the y-axis. Within the chromatogram are peaks, and each peak indicates a different analyte that was separated in the mixture. From each peak, a mass spectrum can be obtained. The mass spectrum is a plot of the molecular ion and the fragmented ions where m/z is on the x-axis and abundance of the 39 Negatively charged / surface Electrons ejected upon imp act To ground through amp lifier Figure 2.5: Diagram of continuous-dynode electron multiplier. 40 peaks is on the y-axis. From each mass spectrum, the analyte can be identified since the fragmentation pattern under the same conditions is specific to that analyte. 2.2 Data Pretreatment Procedures Before subjecting the chromatograms to data analysis, data pretreatment procedures can be applied in order to eliminate non-chemical sources of variance, such as variation in injection volume as well as instrumental drift. The data pre-treatrnent procedures investigated in this research were smoothing, retention time alignment, and normalization. 2. 2.1 Savitzky-Golay Smoothing Before retention time alignment and normalization of the data set, the chromatograms can be smoothed in order to reduce the instrumental noise introduced during analysis. The algorithm used in this research is known as the Savitzky-Golay smooth [2], which is a least squares polynomial smooth. The algorithm reduces the noise in the chromatogram through recalculating each data point in the experimental chromatogram using a polynomial, which is fitted to a section at a time. Before smoothing, two parameters must be defined by the user: order of the polynomial and window size. The order of the polynomial defines the equation which will be used to fit each section. A second or third order polynomial is usually selected since the graphs of these polynomials are similar to the peak shape in a chromatogram [3]. With smaller or larger orders, the peaks in the chromatogram may over-smoothed, 41 thus broadening them. The window size is the number of data points within the chromatogram to which the polynomial is fitted. A window size less than the number of data points contained within a peak will usually provide the best smooth for the chromatogram. However, an extremely small window size will not remove much of the noise in the chromatogram. With a large window size, the peaks in the chromatograms will be over-smoothed, thus reducing the peak height and broadening the peak width. An odd numbered window size is always chosen since the algorithm will average the two data points in the middle of an even numbered window and replace those two points with only one. From this, the number of data points is decreased, thus reducing the chemical information from that chromatogram. Once the two parameters are determined, the polynomial is fitted to the window at the beginning of the chromatogram. This is done by the algorithm solving the polynomial and recalculating the data point in the middle of that window. Next, the new data point replaces the experimental one, thus smoothing that point. The process repeats one data point at a time until each data point is smoothed. Although the Savitzky-Golay smooth reduces the noise in the chromatogram, over-smoothing can occur due to a polynomial order that does not fit the chromatographic peak shape or a large window size. Also, the algorithm does not smooth the first and last data points in a chromatogram. For example, with a window size of 15, the first 7 and the last 7 points in a chromatogram will not be smoothed. Thus, not all of the data points in the chromatogram will be smoothed. 42 2. 2.2 Retention Time Alignment Retention time alignment corrects for minor shifts in the peaks within a chromatogram due to slight variation in the stationary phase and flow rate within the instrument. Multiple retention time alignment algorithms are available, but only two, the peak matching algorithm and the correlation optimized warping (COW) algorithm, are investigated in this research. Along with selecting the alignment algorithm, a target chromatogram to which the peaks are aligned in the sample chromatograms must be chosen. 2. 2. 2. 1 Target Selection Multiple options are available for the target chromatogram. One of the more common options is to use a sample chromatogram fi'om the data set which has many similar peaks to the other chromatograms [4]. The limitation of this selection is that the chromatograms in the data set may not contain all of the peaks necessary for alignment. Another option is a consensus target. The consensus target is a combination of all of the mixtures in the data set, which is analyzed in a same manner as the other mixtures. This is a better choice for a target, since all of the peaks present in the sample chromatograms are present in the target chromatogram. However, the peaks in the consensus target should not be at a low abundance and must be detectable by the algorithm in the alignment procedure. When a consensus target is not possible to form due to variation in chemical components among mixtures, an average target is also applicable to align the data set. An average target is an average of each data point at each retention time within a set of chromatograms. This average chromatogram contains all of the peaks from the 43 other chromatograms. As with the consensus target, the abundance of the peaks in the target chromatogram may not be at a high enough abundance for alignment of the sample chromato grams. 2. 2. 2. 2 Peak Matching Algorithm One of the alignment algorithms investigated in this research was the peak matching algorithm [4]. The algorithm identifies and aligns the peaks in the sample chromatograms and the target chromatogram through determining the zero crossing of the first derivative of each peak. The only parameter besides the target chromatogram that needs to be specified is the window size. The window size determines the number of data points that the zero crossing can be shifted in order to align a peak. With a small window size, such as 1 or 2, the algorithm may not be able to shift the peak in order to align them. However, large window sizes, such as 7 or 8, could align a peak within the sample chromatogram to a different peak in the target chromatogram. In order to align the chromatograms, the peaks in the target chromatogram are identified through initially calculating the first derivative of the entire chromatogram. Then, the algorithm identifies the leading edge of a peak when the standard deviation surpasses a threshold of five times the standard deviation of the noise. The standard deviation of the noise is determined by a user defined number of points in the chromatogram. The algorithm then identifies the apex and the tailing edge of the peak through the zero crossing of the first derivative. The peak is then added to a list of peaks recognized by the algorithm for that chromatogram. Once all the peaks in the target chromatogram have been determined, the algorithm repeats the process for the sample 44 chromatograms. With all of the peaks within each chromatogram having been identified, the peaks in each sample chromatogram are compared to the peaks in the target chromatogram. In order to align a peak, the peak must be present in both the sample chromatogram and the target chromatogram and also be within the allotted window size. If these conditions are met, the peaks from the sample chromatogram and target chromatogram will be aligned through interpolation of data points so that the retention times of the two peaks are the same. If a peak is present in the target chromatogram and not present in the sample chromatogram, the peak is disregarded, and the algorithm moves to the next peak. This process is repeated for each peak in each of the sample chromatograms. The peak matching algorithm can successfully align complex chromatograms, such as diesel samples [4]. However, a peak with a low signal-to-noise ratio may not be identified or aligned by the algorithm since it may be below the baseline noise. Also, the algorithm assumes that the closest peak in the sample chromatogram within a given window size should be aligned to the peak in the target chromatogram. Another disadvantage would be that the algorithm may not be able to align peaks that coelute, especially if one peak is substantially more abundant than the other. 2. 2. 2.3 Correlation Optimized Warping Algorithm The correlation optimized warping (COW) algorithm optimizes correlation between the sample chromatogram and the target chromatogram in order to align the peaks within each chromatogram [5,6]. The algorithm divides the chromatogram into segments, and through adding or subtracting points, assesses correlation between these 45 segments in order to determine the best alignment between the sample and target chromatograms. Along with selection of the target chromatogram, two other parameters must be determined: the segment size and the warp size. The segment size is the number of data points contained within a segment of the chromatogram. When choosing a segment size, the segments formed must have at least the number of data points that define one peak. With a large segment size, such as 75, the segment could contain multiple peaks which would hinder alignment of individual peaks. With small segment sizes, such as 25, the peaks may be divided such that the apex of a peak could be separated from the peak edge. The warp size is the number of points that can be added or subtracted from each defined segment in order to align the peaks within the sample chromatogram to the target chromatogram. With a large warp size, such as six, the segment may be shifted further from the corresponding peaks in the sample chromatograms which would increase the number of misalignments. For a small warp size, such as two, the segment may not be shifted enough to align the sample chromatogram with the target chromatogram. Once these parameters are chosen, the algorithm begins at the end of the chromatograms. For the first segment, the algorithm optimizes correlation between the target and sample chromatograms. This is done through adding and subtracting the warp size from the segment being aligned. With a warp size of n, n data points can be added or subtracted from that segment. For example, the algorithm can add or subtract one, two, or zero data points with a warp size of two. For each addition or subtraction, the algorithm interpolates points so that the target segment and the sample segment are the same length. Once the points have been interpolated, a local correlation coefficient between the sample 46 segment and the target segment is calculated for each warp performed. So for a warp size of two, five local correlation coefficients have been calculated. Then, the algorithm proceeds to the next segment and repeats the process. Each segment is aligned in the same manner until all local correlations for each segment have been calculated. The local correlation coefficients from each segment are summed in all combinations in order to form global correlation coefficients. The best alignment between the sample chromatogram and the target chromatogram is determined by the global correlation coefficient with the highest value. The process is the repeated for the other sample chromatograms. As with the peak matching algorithm, the COW algorithm can be used to align complex chromatograms. Furthermore, the COW algorithm will still align chromatograms even if they differ in the number of data points, noise level, and baseline drift [5]. However, the COW algorithm can align the fronting or tailing edge of the peaks instead of the apices in order to optimize correlation between the target segment and the sample segment. 2. 2.3 Normalization Due to variation in injection volume, the raw chromatograms obtained vary in abundance of peaks among replicates. Through normalization, the variation in abundance observed can be reduced or eliminated. Multiple normalization procedures are available; however, a combination of the maximum peak area normalization followed by the total area normalization of each set of replicates is used for this research. The two norrnalizations are used to correct for differences in peak abundance among the samples 47 in the data set and also to correct for variation in abundance among replicates of those samples. The maximum peak area normalization is first applied to the chromatograms in order to scale all of the chromatograms within the data set to each other. Using this normalization, the larger peaks in the chromatograms are similar in abundance; however, the smaller peaks still vary in abundance among replicates. Total area normalization of the replicates is used scale the smaller peaks in the chromatograms to a similar abundance. Due to the variation in the number of chemical components among samples, total area normalization could not be applied to the entire data set. 2.3 Principal Components Analysis Principal components analysis (PCA) is a multivariate statistical technique that reduces the number of variables in a data set to those that contribute most to the variance [7]. This is done in order to determine the variables that can differentiate the samples from each other. Through PCA, the association and discrimination of samples can be assessed from a few variables rather than the entire data set. In order to apply PCA, the data set must be mean-centered in order to ensure that the first principal component (PC) describes the maximum variance within the data set. In PCA, each data point in each sample is considered a variable. In order to mean-center the data, the average across each variable is calculated. Then, the average is subtracted from each data point in the data set. The process is repeated until every data point in each sample has been mean-centered. 48 Once the data is mean-centered, the covariance matrix is calculated. The covariance is a measure of the amount of variation from the mean between a pair of samples. To form the covariance matrix, the covariance is calculated between all variables, and the values form an n x n matrix, where n is the number of variables in the data set. Within the covariance matrix, the top to bottom diagonal is the covariance of the variable by itself. About that diagonal, the matrix is symmetric since the covariance of variables a x b is the same as the covariance of the variables b x a. From the covariance matrix, the eigenvectors and eigenvalues can be calculated. An eigenvector is a solution vector to the determinant of an n x n matrix. An eigenvalue is a value that when multiplied by the eigenvector and the n x n matrix, the resulting solutions are equal to each other. In order to calculate the eigenvalues, the determinant of the covariance matrix is set equal to zero and then solved. Once an eigenvalue has been determined, the eigenvector can be solved using both the covariance matrix and that eigenvalue. Other eigenvectors are calculated in the same manner and positioned orthogonally to the preceding eigenvector until n-l eigenvectors have been calculated, where n is the number of samples in the data set. Each eigenvector calculated is considered a principal component (PC), and each PC has a corresponding eigenvalue. In order to determine which PC varies the most within a data set, the eigenvalues are ordered from highest to lowest. The PC with the highest eigenvalue is considered the first principal component, since it accounts for the most variance in the data set. From the PCs, the scores for each sample in the data set can be calculated using the mean-centered data. For each PC, each variable from the mean-centered data is multiplied by the corresponding data point in that PC. Then, the resulting values are 49 summed in order to obtain the score for that sample. The calculated score represents the positioning of the sample on a specific PC. These calculations are repeated with each sample in the data set and with each calculated PC. From the calculated scores, eigenvectors, and eigenvalues, the scores plots and the loadings plots can be generated to visually assess the differences in the data set. Typically, the scores for each sample on the first two PCs are plotted to form the scores plot. From the scores plot, the similarities and differences among the samples on those two PCs can be visually observed. Samples that are positioned closely in the scores plot are similar to each other on those two PCs while samples that are further apart are different. The loadings plots are formed by plotting the PC versus the original x-variable, such as retention times of chromatographic data. From the loadings plot, the variables that vary the most within that PC can be identified. Furthermore, the positioning of the samples on the scores plot can be explained using the loadings plot, since the variables present in the loadings plot are the ones that differ the most among samples. The variables that contribute most to the variance in a data set can be established using PCA. However, since only the PCs with the highest percentage of variance are usually assessed, some of the variables which contribute to the variance in the data set are disregarded. Also, the scores plot is visually assessed, thus introducing subjectivity in interpretation of the association and discrimination of samples. 50 2.4 Pearson Product Moment Correlation Coefficients Pearson product moment correlation (PPMC) coefficients assess the correlation between two samples [8, 9]. In order to determine the PPMC coefficient (r) between a pair of samples that are defined as x and y, the n data points within each sample are subjected to the following Equation 2.1: __ Z?=1(xi-f)(J'i-J7) rxy — Jzil=1(xi-f)z'zil=1(3’i’y)2 Equation 2.1 From this equation, correlations between -1 and +1 can be calculated. A perfect inverse correlation is indicated by -1 where the $10pe of one sample is increasing while the slope of other sample is decreasing. A perfect positive correlation of +1 indicates that both samples are increasing or decreasing with the same slope. No correlation between the samples is indicated by zero. From the PPMC coefficient calculated, varying degrees of similarity can be determined. A strong correlation is indicated by PPMC coefficients between 0.8 and 1; a moderate correlation is indicated by PPMC coefficients between 0.5 and 0.8, while PPMC coefficients less than 0.5 indicate a weak correlation [8,9]. When PPMC coefficients are calculated, all of the data points within each sample are compared. Furthermore, a numerical value is calculated using PPMC coefficients. However, the similarity of the samples in the entire data set to each other cannot be assessed, since only two samples can be compared using PPMC coefficients. 51 2.5 Hierarchical Cluster Analysis Hierarchical cluster analysis (HCA) is a multivariate statistical procedure where the similarity of samples is determined through distance among samples in a multidimensional space [10]. With HCA, there are two types of grouping methods: agglomerative and divisive. In the agglomerative method, each sample is defined as its own group, and the samples are clustered together until all of the scores are contained into one group. With the divisive method, all of the samples are grouped together initially, and the samples are then divided until each sample forms its own group. In order to group samples, the distance between pairs must be calculated. Of the equations available to calculate distance between the samples, the Euclidean distance is commonly used. The Euclidean distance is calculated using Equation 2.2. 1/2 dab = [271(xaj " xbjlzj Equation 2.2 The distance between the two samples is represented by dab- The first sample is represented by xa, while the second sample is represented by xb. The j to m represents the dimensions in the coordinate plane in which the samples are plotted. Once the distances among all pairs of samples are calculated, the samples that have the shortest distance between them are linked. In order to link the samples, a linkage method must be selected. Of the many linkage methods available, the single linkage method and the complete linkage method are common methods. The single linkage method is also known as the nearest neighbor linkage method and is illustrated in Figure 2.6A. In this example, sample 3 is positioned between cluster 1 and cluster 2. The 52 distance between sample 3 and the sample in cluster 1 that is closest to sample 3 is calculated. Similarly, the distance between sample 3 and the sample in cluster 2 that is closest to sample 3 is calculated. Sample 3 is assigned to the cluster to which it has the shortest distance, which in this example is cluster 1. The complete linkage method, also known as the farthest neighbor linkage method, connects the sample with the farthest neighbor closest to the sample. For example, sample 3 is positioned between cluster 1 and cluster 2. When the distances are calculated, sample 3 is observed to be closer to the farthest sample in cluster 2 than the farthest sample in cluster 1. So, sample 3 is linked to cluster 2 (Figure 2.6B). Once the samples with the shortest distance between them have been linked, the distances among the new cluster and the samples are recalculated. Again, the samples with the shortest distance between them are linked. This is repeated until all samples are clustered into a single group. After the samples have been linked, the similarity of the samples to each other can be established using the distances calculated (Equation 2.3). dab dmax Equation 2.3 similarityab = 1 — The distance between two samples is represented by dab: and the greatest distance between samples in the data set is represented by dmax. A similarity of l is assigned to samples that are identical, while a similarity of zero indicates the most dissimilar samples in the data set. The output of HCA is a dendrogram, which shows the calculated similarities of the samples in the data set. On the x-axis of the dendrogram, the similarity from 1 to O is 53 0.. .0 0. 0'2 0 ‘ o o 0:. Cl Figure 2.6: A) Diagram depicting the single linkage method used in HCA. B) Diagram depicting the complete linkage method used in HCA. 54 shown. The y-axis contains all of the samples in the data set. Within the dendrogram, samples that are similar will have a high similarity and be more closely linked to each other. Within the dendrogram each sample begins with its own branch, and samples are then connected according to the similarity calculated previously from highest similarity to lowest similarity. Hierarchical cluster analysis shows the natural clusters of samples through assessing their similarity. Furthermore, a number which measures the similarity among samples is specified. However, HCA cannot determine which variables in the samples contribute to the similarity. With PCA, the differences among samples are determined through the variables that vary the most within a data set. Through HCA, the similarities among samples are assessed, and samples are grouped by their similarity. However, HCA cannot determine which variables contribute the most to the differences in samples. Furthermore, in PCA, each PC only accounts for a certain percentage of the variance and only distinguishes the samples based on those variables. 55 2.6 10. References Skoog DA, Holler FJ, Crouch SR. Principles of instrumental analysis. 6th ed. Belmont, CA: Thomson, 2007. Savitsky A, Golay MJE. Smoothing and differentiation of data by simplified least squares procedures. Anal Chem 1964;36:1627-1639. Beebe KR, Pell RJ, Seasholtz MB. Preprocessing the samples. In: Chemometrics: a practical guide. New York: Wiley, 1998; 26-55. Johnson KJ, Wright BW, Jarman KH, Synovec RE. High-speed peak matching algorithm for retention time alignment of gas chromatographic data for chemometric analysis. J Chromatogr A 2003;996:141-55. Nielsen NPV, Carstensen JM, Smedsgaard J. Aligning of single and multiple wavelength chromatographic profiles for chemometric data analysis using correlation optimized warping. J Chromatogr A 1998; 805: 17-35. Tomasi G, van den Berg F, Andersson C. Correlation optimized warping an dynamic time warping as preprocessing methods for chromatographic data. J Chemometr 2004; 18:231-241. Smith, LI. A tutorial on principal components analysis. 2002. Brereton, RG. Applied chemometrics for scientists. Hoboken, NJ: John Wiley & Sons, Ltd, 2007. Miller JN, Miller JC. Statistics and chemometrics for analytical chemistry. 4th ed. Harlow, England: Pearson Education Limited, 2000. Infometrix, Inc. Pirouette User Manual Version 4.0. Bothell, WA: 2008. 56 CHAPTER 3 Association of Evaporated Ignitable Liquids to Their Neat Counterparts Using Chemometric Procedures 3.1 Introduction At suspected arson scenes, an ignitable liquid may be present at various levels of evaporation depending on the extent of the fire. Due to evaporation of the ignitable liquids during the fire, the more volatile compounds are lost, changing not only the abundance of the compounds, but also the number of compounds detected by gas chromatography-mass spectrometry (GC-MS). Since the overall profile of the chromatogram has changed, an arson investigator may not be able to identify the evaporated liquid used in the fire. In this research, chemometric procedures, such as principal components analysis (PCA), Pearson product moment correlation (PPMC) coefficients, and hierarchical cluster analysis (HCA), were used as an objective method to investigate the association of an evaporated ignitable liquid to its neat counterpart. Firstly, six ignitable liquids, one from each of six different American Society of Testing and Materials (ASTM) classes [1], were evaporated to four levels of evaporation. The neat ignitable liquids along with the corresponding evaporated liquids were analyzed by GC-MS. The total ion chromatograms of the neat and evaporated liquids were compiled. The chromatograms were subjected to pretreatment procedures where the chromatograms were smoothed, retention time aligned, and normalized. Principal components analysis was used to discriminate the ignitable liquids in different classes, 57 while associating the evaporated, ignitable liquids to the corresponding neat ignitable liquid. The PPMC coefficients were used to assess both the precision of the extraction procedure as well as the degree of correlation of the evaporated ignitable liquids to the neat ignitable liquid. Hierarchical cluster analysis was used as a statistical measure to assess the similarity of the positioning of the evaporated liquid to the corresponding neat liquids in the scores plot. 3.2 Materials and Methods 3. 2.1 Sample Collection A subset of six ignitable liquids, one liquid from each of the six classes defrned by ASTM, was collected from various locations. Gasoline and kerosene were collected from service stations in the Lansing, Michigan area. Lacquer thinner, lamp fuel, marine fuel stabilizer, and paint thinner were collected from local hardware stores, grocery stores, and online sources. The ignitable liquids in the subset, along with the class and major compounds present in each, are listed in Table 3.1. 3. 2. 2 Sample Preparation Each of six ignitable liquids was evaporated to four different levels of evaporation (5%, 10%, 20%, and 50% by volume) using filtered air and magnetic stir bars for agitation. Once evaporated, liquids were stored in 7.4 mL amber screw cap vials (Fisher Scientific, Pittsburgh, PA) that were acid washed before use. Vials were labeled, wrapped in Parafilm® (American National Can, Menasha, WI), and stored at 16°C until analysis. 58 Table 3.1 Ignitable liquids investigated. Ignitable Liquid ASTM Class Major Compounds Gasoline Gasoline Toluene, Alkylbenzenes, Naphthalenes Kerosene Petroleum Distillate Alkylbenzenes, Normal Lacquer Thinner Lamp Fuel Marine Fuel Stabilizer Paint Thinner Aromatic Alkane Naphthenic Paraffinic lsoparaffinic Alkanes (Cg-€13) Toluene Normal Alkanes (Crz-Crs) Naphthalenes, Branched Alkanes (C10, C11) Cyclic Alkanes (C6) Branched alkanes (C7, C8, C10) 59 All liquids (neat and evaporated) were diluted in dichloromethane (CHzClz) (spectrophotometric grade, Jade Scientific, Canton, MI). The liquids had different dilution factors so that the same order of magnitude for the abundance was observed in all chromatograms. Paint thinner and marine fuel stabilizer were diluted 1:350 (v/v), while gasoline was diluted 1:100 (v/v), kerosene was diluted 1:10 (v/v), lacquer thinner and lamp fuel were diluted 1:550 (v/v). A 250 1.1L aliquot of each diluted liquid was spiked onto a 5 x 5 cm2 Kimwipe TM (Kimberly-Clark, Irving, TX) in a 1.0 L, unlined paint can (Arrowhead Forensics, Lenexa, KS). The liquids were then extracted in triplicate at 80°C for four hours using a passive headspace extraction with one fourth of an activated carbon strip (Albrayco Technologies, Inc. Cromwell, CT). After extraction, the activated carbon strip was eluted with 200 14L CHzClz and analyzed by GC-MS. A consensus standard was also prepared by combining 250 11L of each of the six neat liquids, diluted as described previously. The consensus standard was then extracted in triplicate and analyzed as previously described. 3. 2.3 GC-MS Analysis All extracts were analyzed using an Agilent 6890 gas chromatograph coupled to an Agilent 5975 mass spectrometer, with an Agilent 7683B automated liquid sampler (ALS) (Agilent Technologies, Santa Clara, CA). The GC was equipped with an Agilent HP-SMS capillary column (30 m x 0.25 mm internal diameter x 0.25 pm film thickness (d0). A 1 11L volume of sample was injected using the ALS in the pulsed, splitless mode, using a pressure pulse of 15.0 psi for 0.25 min. The inlet temperature was 250°C. The 60 carrier gas was ultra-high purity helium (Airgas, East Lansing, MI) and had a nominal flow rate of 1 mL/min. The GC temperature program was as follows: 40°C for 3 min, 10°C/min to 280°C, hold for 4 min at 280°C. The transfer line was maintained at 280°C, and the mass spectrometer was operated in electron ionization mode (70 eV) with a quadrupole mass analyzer operating in full scan mode (m/z 50-550) at a scan rate of 2.91 scans/s. Before any extracts were analyzed, two equilibration runs were performed to minimize active sites in the inlet and on the column. For the equilibration runs, all of the instrument parameters were the same except the oven ramp rate, which was increased from 10°C/min to 20°C/min. The equilibration runs were always aliquots of the first sample to be analyzed on that day. 3. 2.4 Data Pretreatment Total ion chromatograms (TICs) were generated for all the neat and evaporated liquids. All of the chromatograms were smoothed using a Savitzky-Golay smooth in ChemStation sofiware (version E.01.01.335, Agilent Technologies), then compiled into a data set. Chromatograms were retention time aligned to a target chromatogram, which was generated by smoothing and averaging the triplicates of the consensus standard. Both a commercially available correlation optimized warping (COW) algorithm (LineUp TM, version 3.0, Infometrix, Inc., Bothwell, WA) and a peak matching algorithm available in the literature [2] were used to align the chromatograms. For each algorithm, various user-defined parameters were investigated as shown in Table 3.2. There are 61 Table 3.2 User defined parameters investigated for peak matching algorithm and COW algorithm. Peak Matching . Algorithm COW Algonthm Window Size Warp Size Segment Size in Data Points in Data Points in Data Points Data Points 2, 3, 4, 5, 6, 7 1, 2, 3, 4’ 5, 6 15, 25, 45, 60, 65, 75 62 typically 13 points across a peak so that window size and warp size are smaller than the peak width, while the segment size is larger than the peak width. For the COW algorithm, all combinations of the warp sizes and the segment sizes in Table 3.2 were investigated. In order to determine the optimal alignment, chromatograms were visually assessed for peak misalignments. The PPMC coefficients were also calculated in Matlab (version 7.7.0.471, The MathWorks, Inc., Natick, MA) and used to further evaluate the alignment of the chromatograms. After retention time alignment, the chromatograms were normalized using Microsoft Excel (version 12.0.6524.5003, Microsoft Corp., Redmond, WA). Maximum peak normalization was first performed, whereby each data point in a chromatogram was divided by the abundance of the maximum peak and then multiplied by the average abundance of the maximum peaks for all chromatograms. Total area normalization was then performed among replicates for each neat and evaporated ignitable liquid. Each data point in the chromatogram was divided by the total area of that chromatogram and then multiplied by the average area of the replicate chromatograms for that liquid. This combination of normalization procedures reduced the differences in abundance observed within the chromatograms. 3. 2.5 Data Analysis Once the chromatograms had been aligned and normalized, PCA was initially performed on the neat liquids using MatLab (The Math Works, Inc.). The scores for the first principal component (PCI) and the second principal component (PC2) for each neat liquid were calculated in Matlab (The Math Works, Inc.) and plotted in Excel (Microsoft 63 Corp.) to generate a scores plot. The eigenvectors for PCI and PC2 were also plotted against retention time in Excel to generate loadings plots. These plots were used to identify those compounds that were most variable and hence, most discriminating, among the liquids. Scores for the evaporated liquids were then calculated in Excel (Microsoft Corp.). Prior to projection, chromatograms were mean-centered. To mean-center the chromatograms of the evaporated liquids, the abundance at each retention time was subtracted from the average abundance at the same retention time of the neat liquids. The mean—centered data were then multiplied by the eigenvector of PCI and summed to obtain the score for the evaporated liquid on PCI. Scores for each evaporated liquid on PC2 were calculated in a similar manner using the eigenvector of PC2. The scores for the evaporated liquids were then projected onto the scores plot of the neat liquids. Using this approach, association and discrimination was based on chemical composition of the neat liquids rather than the evaporated liquids. Pearson product moment correlation coefficients (Equation 2.1) were calculated for the aligned chromatograms using MatLab (The Math Works, Inc.). The PPMC coefficients were calculated among replicates of each liquid to assess the precision of the extraction and analysis procedure. The correlation between all levels of evaporation and the corresponding neat liquid was then calculated to assess the degree of similarity between the evaporated and neat liquids. Using the scores calculated for the neat and evaporated liquids, hierarchical cluster analysis was performed in Pirouette (version 4.0, Infometrix, Inc., Bothwell, WA). The Euclidian distance (Equation 2.2) was calculated between all pair-wise combinations 64 of the PC scores. Using the agglomerative method, the scores were linked through the nearest neighbor linkage method. Using these parameters, two dendrograms were generated: a dendrogram of only the neat liquids and a dendrogram of the neat and evaporated liquids, from which similarity was assessed. 3.3 Results and Discussion 3. 3.1 Optimization of Retention Time Alignment With the peak matching algorithm, window sizes of two to seven data points were investigated. With the larger window sizes, specifically five to seven, more misalignments were observed among the neat and evaporated liquids. The most misalignments were observed using a window size of seven, since the peaks were shifted further with a larger window size. The chromatograms of marine fuel stabilizer showed major misalignments, especially with a window size of seven (Figure 3.1A). When window sizes of two to four were investigated, the number of misalignments decreased slightly. Of the remaining window sizes, the fewest misalignments among replicates of each liquid were observed using a window size of three (Figure 3.1B). Thus, for the peak matching algorithm, the optimal window size was three. The optimal alignment for the COW algorithm was determined by investigating multiple warp sizes and segment sizes. For most combinations of warp size and segment size, misalignments were still present among the chromatograms of the neat and evaporated liquids. The most misalignments were observed with a large warp size and a large segment size. With large segment sizes, each segment contains multiple peaks. 65 ”eufiaaw 1 1 .6 Retention Time (min) 1 l .8 Figure 3.1: A) Poorly aligned 2,6-dimethylundecane peak in the TIC of neat and evaporated marine fuel stabilizer using a window size of seven. B) Well aligned 2,6- dirnethylundecane peak in the TIC of neat and evaporated marine firel stabilizer using a window size of three. 66 When a large warp size is also used, these large segments are shifted further, which results in misalignment of the peaks. For example, a warp size of six and a segment size of 45 showed major misalignments for the 2,2,3,5-tetramethylheptane peak in paint thinner (Figure 3.2A). Visual assessment of the chromatograms yielded optimal parameters of a warp size of three and a segment size of 65 for the COW alignment (Figure 3.2B). The optimal COW alignment and the optimal peak matching alignment were then compared to determine the optimal alignment method. The COW alignment was observed to have fewer major misalignments than the peak matching alignment when chromatograms were compared visually (Figure 3.3A and B). Higher PPMC coefficients were also observed among the replicates of the COW aligned chromatograms than the peak matching aligned chromatograms (Table 3.3). For each ignitable liquid, a two sample t-test was applied to compare the mean PPMC coefficients calculated for each alignment algorithm. The difference between the mean PPMC coefficients was found to be statistically significant at the 90% confidence level for all ignitable liquids except kerosene. Thus, the significantly higher PPMC coefficients indicated improved alignment with the COW algorithm compared to the peak matching algorithm. 67 ...fiél’fldflfi 0 ’ -- * - ~- . 7 .5 Retention Time (min) 7.6 6.0E05 B 3 2: <2 7.5 Retention Time (min) 7.6 Figure 3.2: A) Poorly aligned 2,2,3,5-tetramethylheptane peak in the TIC of neat and evaporated paint thinner with a warp size of six and a segment size of 45. B) Well aligned 2,2,3,5-tetramethylheptane peak in the TIC of neat and evaporated paint thinner with a warp size of three and a segment size of 65. 68 6.0E05 A Abundance 7.48 Retention Time (min) 7.58 6.0E05 B 4999929“. . 7.5 Retention Time (min) 7.6 Figure 3.3: A) Poorly aligned 2,2,3,5-tetramethylheptane peak in the TIC of neat and evaporated paint thinner with a window size of three using the peak matching algorithm. B) Well aligned 2,2,3,5-tetramethylheptane peak in the TIC of neat and evaporated paint thinner with a warp size of three and a segment size of 65 using the COW algorithm. 69 Table 3.3: PPMC coefficients for all replicates (n=15) for optimal peak matching alignment and optimal COW alignment. Mean PPMC Coefficient Mean PPMC Coefficient :1: :1: Standard Deviation Standard Deviation Ignitable quurd Using COW Algorithm Using Peak Matching (Warp 3, Segment 65) Algorithm (Window Size 3) Gasoline 0.9951 2 0.0044 0.9915 :i: 0.0042 Kerosene 0.9958 d: 0.0016 0.9954 :t 0.0018 Lacquer Thinner 0.9936 2 0.0090 0.9822 i 0.0178 Lamp Fuel 0.9964 :‘c 0.0018 0.9895 2 0.0102 Marin.“ Fm 0.9972 i 0.0018 09959 :t 0.0021 Stabrlrzer Paint Thinner 0.9949 2 0.0044 0.9892 :t 0.0109 70 Combinations of the peak matching algorithm and the COW algorithm were also considered. However, preliminary investigation of the combination of the algorithms showed an increased number of misalignments in the chromatograms, and this approach was not further investigated. Thus, through visual assessment of chromatograms and PPMC coefficients, the COW alignment with a warp of three and a segment size of 65 was chosen as the optimal alignment. 3. 3.2 Normalization The unnormalized chromatograms showed differences in abundance among triplicate extractions for both evaporated and neat liquids (Figure 3.4A). Maximum peak area normalization was first performed on the chromatograms, where the replicates were generally grouped by evaporation level, as illustrated for marine fuel stabilizer in Figure 3.43. After maximum peak normalization, there was still some spread in the replicates. Total area normalization between each set of replicates was then performed in order to minimize these differences (Figure 3.4C). Through a combination of maximum peak area normalization and total area normalization of the replicates, differences in abundance of the triplicate extractions of the evaporated liquids were minimized. 3. 3.3 Association and Discrimination of the Neat Liquids using PCA The scores plot of the first principal component (PCI) and the second principal component (PC2) accounts for 67.5% of the variance among the neat liquids (Figure 3.5). As observed in Table 3.4, there is at least an order of magnitude difference between the 71 r3505 A Abundance Abundance 10.48 Retention Time (min) 10.56 4 0507 ( 8 5 'U S ..O < _-/" t f I ., I fixinwf, 0 l 10.56 1 0.48 Retention Time (min) Figure 3.4: A) Unnormalized pentylcyclohexane peak in the TIC of neat and evaporated marine fuel stabilizer. B) Maximum peak normalized pentylcyclohexane peak in the TIC of neat and evaporated marine fuel stabilizer. C) Total area normalization of replicates of the pentylcyclohexane peak in the TIC of neat and evaporated marine fuel stabilizer. Neat (—), 5% Evaporated (— u), 10% Evaporated (- u u), 20% Evaporated (— - ), 50% Evaporated (- u u ). 72 2.5E08 O 0 $3 0. 8, 0 8 a, A O at o -2.5E08 * 4.0E08 0 4.0E08 PG] (44.5%) Figure 3.5: Scores plot of PC] vs. PC2 based on the TIC for the six neat ignitable liquids. Liquids were indicated by symbol. Lamp Fuel (I), Kerosene (0), Marine Fuel Stabilizer (0), Paint Thinner (O), Gasoline (A), and Lacquer Thinner (t). 73 Table 3.4: Mean PCA score for replicates (n=3) of each liquid class based on the TIC. Mean PCA score on PCI Mean PCA score on PC2 Ignitable Liquid Standard Seviation for Standard Dteviation for Replicates Re licates Paint Thinner -1 .48x108 i 3.74x105 1.56x10 i 4.68x105 Marine Fuel Stabilizer 1.24x108 a 2.66x106 1.78x108 2 3.91x106 Kerosene 3.24x108 a: 52le06 -4.74x107 2 5.64x106 Lamp Fuel -2.25x10° :t 8.75x105 -1.80x108 2h 2.25x106 Lacquer Thinner -1 .43x108 2: 1.29x106 -6.24x107 2 3.91x105 Gasoline -1.56x108 :1: 3.64x106 -4.39x107 :1: 1.93x106 74 average and the standard deviation for the scores of the ignitable liquids on PCI and PC2. The difference in order of magnitude indicates that the replicates of each ignitable liquid are not spread from each other in the scores plot. Very little spread is observed among the replicates of each of the liquids, indicating a precise analytical procedure. Overall, the liquids are differentiated in PC] and PC2, except for gasoline and lacquer thinner. Kerosene and marine fuel stabilizer load positively on PC], while gasoline, lacquer thinner, lamp fuel, and paint thinner load negatively. In PC2, marine fuel stabilizer and paint thinner load positively, while gasoline, kerosene, lacquer thinner, and lamp fuel load negatively. The loadings plots for PCI and PC2 are shown in Figure 3.6. The first principal component (Figure 3.6A) discriminates the ignitable liquids based on toluene, Cz- and C3-alkylbenzenes, and branched alkanes (C7, C3, and C10), which load negatively on PCI and naphthalenes, cyclic alkanes, and normal alkanes (C9-C13) that load positively. The second principal component (Figure 3.6B) discriminates the ignitable liquids based on toluene, Cz- and C3-alkylbenzenes, and normal alkanes (C12-C17) that load negatively and branched alkanes, cyclic alkanes, and naphthalenes that load positively. The loadings plots of PC] and PC2 can be used to explain the positioning of the liquids in the scores plot based on chemical composition. Paint thinner, which contains only C7, C8, and C10 branched alkanes (Appendix A, Figure A.4), is positioned negatively on PCI and positively on PC2. The C7, C3, and C10 branched alkanes are the only compounds contributing to the positioning of paint 75 thinner in the loadings plots. These branched alkanes load negatively on PCI and positively on PC2. Marine fuel stabilizer, which contains naphthalenes, cyclic alkanes, C10 and C11 branched alkanes, and 2,6-dimethylundecane (Appendix A, Figure A3), is positioned positively on PCI due to the presence of naphthalenes and cyclic alkanes that load positively on PCI. Marine fuel stabilizer is positioned positively on PC2 due to the presence of naphthalenes, cyclic alkanes, branched alkanes, and 2,6-dimethylundecane that load positively on PC2. Most of the positive variance described by PC2 is due to the compounds present in marine fuel stabilizer. Kerosene is dominated by normal alkanes (Cm-C19), but also contains naphthalenes and C2- and C3-alkylbenzenes (Appendix A, Figure A.2). In the scores plot, kerosene is positioned positively on PCI and negatively on PC2. The normal alkanes and naphthalenes load positively on PCI while the C2- and C3-alkylbenzenes load negatively on PCI. Since the naphthalenes and the C12-C13 normal alkanes are more dominant than the aromatic compounds, kerosene is positioned positively on PC]. The C12-C17 normal alkanes and the C2- and C3-alkylbenzenes are contributing more to the variance on PC2 than the early eluting naphthalenes, which positions kerosene negatively on PC2. 76 0.20 r A Naphthalenes and Cyclic Alkanes I 5 C16 ‘1‘ ; z i 3 3 _ L ['7 (£18 .301 . I“ -8 fl 3 L___l 3 anCth C7, C8, C10 5 1 -0.20 3 Retention Time (min) 21 Naphthalenes, Cyclic Alkanes, : 7 Branched Alkanes I ' 9 ' 6 8 N U f 9" i l a 0 I r *W H * I T ' :5 1 4 l__J I CI6 17 cu 3 C3-Alkylbenzenes C15 .3 1 C12 C14 4) 26 3 C” ' 3 Retention Time (min) 2] Figure 3.6: Loadings plots of (A) PC] and (B) PC2 based on the TIC for six neat liquids. Major components are labeled: 1) toluene, 2) ethylbenzene, 3) o-xylene, 4) p-xylene, 5) 1-ethyl-3-methylbenzene, 6) 2,2,3,5-tetramethylheptane, 7) 5-ethyl-2,2,3- trimethyheptane, 8) 3,6-dimethylundecane, 9) 2,6-dimethylundecane. 77 Lamp firel, which contains C12-C15 normal alkanes (Figure 3.7A)(Appendix A, Figure A.1), is positioned at zero on PCI. Within the loadings plot, the C12-C15 normal alkanes load positively on that PC (Figure 3.6A). The mean-centered chromatogram of lamp fuel contains both components from lamp fuel and components from other ignitable liquids (Figure 3.7B). These other components are present in the average chromatogram to which the lamp fuel is mean-centered. When multiplied by the eigenvector for PCI, the mean- centered chromatogram for lamp fuel contains similar peak areas loading positively and negatively, which offset one another (Figure 3.7C). Thus, components from the other liquids that load positively and negatively on PCI contribute to the positioning of lamp fuel. Consequently, the scores for lamp filel are positioned at zero on PC]. The C12-C15 normal alkanes load negatively on PC2, which contributes to the negative positioning of the liquids on that PC. Gasoline, which contains toluene and C2- and C3-alkylbenzenes (Appendix A, Figure A.5), is positioned negatively on PC] in the scores plot due to the dominance of these compounds, which load negatively on that PC. Gasoline is positioned slightly negatively on PC2 due to toluene and the C2- and C3-alkylbenzenes which have a small negative contribution to the loadings plot of PC2. 78 ”FWD. mg.-.— —.-r_. ‘ lat-:03 ”K o 8 || 2 § . E ..D . «r: 0 L 3 Retention Time (min) 2 1 801-107 18 O 8 _ .3 ; § '3 1 mi '1'] T f ' V -4.0l507 3 Retention Time (min) 21 I. I Em ....(fu. . ... .. -.... ......-..-.....-........._....._-_...._._.._,_-... _. .1 l 0 I 8 8 5 ..D < . 1 3 Retention Time (min) 21 Figure 3.7: A) TIC of lamp fuel. B) Mean-centered chromatogram of lamp fuel. C) Mean-centered chromatogram multiplied by the eigenvector for PCI. 79 Lacquer thinner, which only contains toluene (Appendix A, Figure A6), is positioned negatively on PC 1 and PC2 since toluene loads negatively on both PCs. Lacquer thinner and gasoline are positioned closely in the scores plot and cannot be visually distinguished. Lacquer thinner contains only toluene, while gasoline mainly contains toluene and the C2- and C3-alkylbenzenes. In the loadings plots for PCI and PC2, toluene and the C2- and C3-alkylbenzenes load negatively on both PC] and PC2. Although gasoline contains more components on PC], toluene is contributing more to the positioning of the liquids on both PC] and PC2 than the C2- and C3-alkylbenzenes. Thus, gasoline and lacquer thinner are positioned similarly in the scores plot. Within the scores plot, the replicates of each liquid are well associated to one another indicating precision in the analytical methodology. With the exception of gasoline and lacquer thinner, the ignitable liquids are separated from one another in the scores plot due to their differences in chemical composition. 3. 3. 4 Association of Evaporated Liquids to Neat Liquids using PCA When the evaporated liquids are projected onto the scores plot, the liquids are positioned closely to the corresponding neat liquid (Figure 3.8). The replicates of each evaporated liquid are also clustered closely together, indicating precision in the analytical methods. Evaporated paint thinners at the 5%, 10%, and 20% evaporation levels (Appendix B, Figures B.13 to B.15) are positioned closely to the neat liquid. Due to evaporation, the 50% evaporated liquid (Appendix B, Figure B.16) contains a lower concentration of early 80 eluting C7, C3, and C10 branched alkanes. As a result, these compounds contribute less to the negative variance of PCI, shifiing the positioning of the 50% evaporated liquid more positive. Similarly, the compounds contribute less to the variance on PC2; hence, the 50% evaporated liquid is positioned less positively on PC2. Evaporated marine fuel stabilizer at the 5% evaporation level is positioned closely to the neat liquid (Appendix B, Figures B9 to 8.12). As evaporation increases further, the evaporated liquids are positioned slightly less positively on both PC] and PC2. This trend is due to the evaporation of early eluting naphthalenes, branched alkanes, and cyclic alkanes, which decrease the positive contribution of these compounds to the variance on PC] and PC2. Evaporated kerosenes (Appendix B, Figures 135 to B8), specifically the 5%, 10%, and 20% evaporated liquids, are closely associated with the neat liquid on both PC] and PC2. At these evaporation levels, only the aromatic components have been evaporated. Since the C2- and C3-alkylbenzenes contribute only slightly to the variance described in the loadings plots, there is little change in positioning of the liquids on the scores plot. At the 50% evaporation level, the C9-C1] normal alkanes and early eluting naphthalenes have been evaporated, along with the aromatic components. Therefore, the C12-C17 normal alkanes, which load negatively on PCI and PC2, are the compounds that contribute to the positioning of the 50% evaporated liquid. Due to the evaporation of the early eluting naphthalenes and the early eluting normal alkanes, the 50% evaporated liquids are positioned closer to zero on PCI and less positively on PC2. 81 2.5E08 PC2 (23.0%) C an 20' -2.5E08 C‘ 4.01308 0 4.0E08 l>Cl (44.5%) Figure 3.8: Scores plot of PCI vs. PC2 based on the TIC for six ignitable liquids with projections of evaporated liquids. Liquids are indicated by symbol. Lamp Fuel (I), Kerosene (0), Marine Fuel Stabilizer (0), Paint Thinner (O), Gasoline (A), and Lacquer Thinner (*). Each fill indicates a different level of evaporation. Neat (Filled e.g. I), 5% evaporated (Half Filled e. g. B), 10% evaporated (Cross e.g. B), 20% evaporated (Line e.g. B), 50% evaporated (No Fill e.g.Cl). 82 Evaporated lamp fuels (Appendix B, Figures B.1 to B4) are closely associated with the neat lamp fuel. However, as evaporation level increases, the liquids are positioned slightly more negatively on PCI and slightly less negatively on PC2. This slight shift in positioning is due to the decrease in contribution of the C12 and C13 normal alkanes due to evaporation, which load positively on PC] and negatively on PC2. Evaporated lacquer thinners (Appendix B, Figures 8.2] to B.24) are positioned closely to the neat lacquer thinner on both PCI and PC2 since compounds that eluted in the solvent front were evaporated. As a result, the abundance of toluene is concentrated by evaporation. After normalization, the toluene abundance is similar for all levels of evaporation and also similar to the levels observed in the neat lacquer thinner. Within the scores plot, all evaporated lacquer thinners overlaid with the neat lacquer thinner. Evaporated gasolines (Appendix B, Figures B.17 to B.20) are very closely associated with the neat liquid and are positioned negatively on both PC] and PC2. Gasoline contains components more volatile than toluene that elute in the solvent front and are not detected by GC-MS. The presence of these components has been confirmed using GC with flame ionization detection. During evaporation, these more volatile components are evaporated, resulting in toluene and C2- and C3—alkylbenzenes being concentrated. However, after normalization, the evaporated and neat gasolines were similar in abundance. Hence, no differences in positioning of the evaporated liquids to the neat gasoline are observed. Thus, the evaporated liquids are closely associated to the corresponding neat liquid, with the exception of 50% evaporated kerosene. The evaporated liquids are 83 differentiated from evaporated liquids in other ASTM classes, with the exception of gasoline and lacquer thinner. Evaporated gasolines and lacquer thinners are positioned closely since both contain toluene, which is contributing greatly to the variance in PCI and PC2. As with the neat liquids, PCA can only be visually assessed, and there is not a statistical measure of the association of the evaporated liquids to the neat liquids. Therefore, PPMC coefficients and HCA were used to statistically evaluate the similarities among the evaporated and neat liquids 3.3.5 Association and Discrimination of Evaporated Liquids to Neat Liquids using PPMC coeflicients The precision of the analytical methods was investigated by calculating mean PPMC coefficients between the neat replicates of each liquid (Table 3.5) (Appendix D, Table D.1). For all neat liquids, the mean PPMC coefficients are greater than 0.99, indicating a strong correlation and hence, acceptable precision in the extraction and analysis procedures. The PPMC coefficients of the replicates are not 1.0000 due to slight instrument variations in the analyses of the replicates and slight variability in the extraction procedure. The PPMC coefficients were also calculated to investigate the chemical similarities between the evaporated ignitable liquids and their neat counterparts, as well as differences between ignitable liquid classes. While each principal component accounts for a certain percentage of the variance, PPMC coefficients take into account the entire chromatogram and are an alternative method for comparison of the ignitable liquids. 84 Table 3.5: PPMC coefficients for replicates (n=3) and range of coefficients for each liquid class based on the TIC. Mean PPMC Coefficient Range of PPMC I i table Li uid :l: Coefficients between gn q Standard Deviation for Neat & Evaporated Neat Replicates Liquids in Each Class Gasoline 0.9944 2 0.0007 0.9994 - 0.9759 Lacquer Thinner 0.9984 :1: 0.0013 1.0000 - 0.9593 Lamp Fuel 0.9953 2: 0.0018 0.9997 - 0.9716 Paint Thinner 0.9920 :t 0.0013 0.9998 - 0.9157 Marin.“ Fuel 0.9982 a 0.0010 0.9994 - 0.8535 Stabrlrzer Kerosene 0.9954 i 0.0006 0.9992 - 0.8744 85 A strong correlation (Section 2.4) is observed between the neat and the evaporated liquids for all six liquids at all levels of evaporation (Table 3.6) (Appendix C, Tables CI to C6). For gasoline, lacquer thinner, lamp fuel, and paint thinner, all PPMC coefficients are greater than 0.9000, which indicates that, even at the 50% evaporation level, the evaporated liquids are still sufficiently similar to yield strong association to the neat liquid. For marine fuel stabilizer, there is a wider range of PPMC coefficients due to the loss of earlier eluting compounds from the chromatogram. However, a strong correlation is still observed between the neat and 50% evaporated liquids. Like marine fuel stabilizer, kerosene shows a wider range of PPMC coefficients which is due to the loss of all volatile compounds at the 50% evaporation level. As a result, the chemical composition of the neat and 50% evaporated kerosene is sufficiently different that a lower PPMC coefficient (0.8764) is observed. However, the correlation between the neat and the 50% evaporated liquid is still considered a strong correlation. Although strong correlations between the evaporated liquids and the corresponding neat liquids are observed, PPMC coefficients for lacquer thinner are lower than expected. The only peak in the lacquer thinner chromatograms is toluene, which is concentrated during evaporation. The wide range of PPMC coefficients for lacquer thinner is due to the misalignment of toluene in the chromatograms even after optimizing alignment. With the low abundance of this peak in the consensus target, the algorithm is unable to optimize the correlation between the consensus target and the sample chromatogram. Thus, misalignments occur in the chromatograms of lacquer thinner. For lamp fuel, the lower correlations observed between the neat and evaporated liquids are due to the evaporation of the C12 and C13 normal alkanes. Although PPMC 86 coefficients are not affected by overall peak height, the change in the relative peak heights will affect the correlation coefficient. As the level of evaporation increases, the peak height of the C12 and C13 normal alkanes decreases in relation to the other components; this decreases the correlation coefficient between the neat and evaporated lamp fuels Neat and evaporated gasolines are positioned closely to lacquer thinner in the scores plot (Figure 3.8) and could not be differentiated by PCA. The PPMC coefficients were calculated between the gasoline and lacquer thinner to assess the degree of similarity between the two liquids (Table 3.6). Neat and 50% evaporated levels of each liquid were chosen to represent the two extremes in chemical composition. When the 50% evaporated liquid was compared to the corresponding neat liquid, a strong correlation was observed. With all other combinations of the neat and 50% evaporated lacquer thinner and the neat and 50% evaporated gasoline, moderate to weak correlations were observed (Table 3.6) (Appendix C, Tables C5 and C6; Appendix D, Tables D5 and D6). Lacquer thinner only contains toluene, while gasoline contains not only toluene, but also the C2- and C3-alkylbenzenes. Hence, moderate to weak correlations are observed between these liquids, irrespective of the level of evaporation, indicating that the liquids can be distinguished using PPMC coefficients. However, the two liquids cannot be differentiated using PCA, because PCI and PC2 only account for a percentage of the variance. Thus, the combination of PCA with PPMC coefficients can differentiate the classes of ignitable liquids from each other. 87 Table 3.6: PPMC coefficients (n=9) between neat and evaporated gasoline and neat and evaporated lacquer thinner. Mean PPMC Coefficient :l: Parr-wrse comparison Standard Deviation Neat Gasoline vs. 50% Evaporated Gasoline 0.9848 :1: 0.0028 Neat Lacquer Thinner vs: 50% Evaporated 0.9978 :1: 0.0031 Lacquer Thinner Neat Gasoline vs. Neat Lacquer Thinner 0.5423 :1: 0.0030 50% Evaporated Gasoline vs. Neat Lacquer Thinner 0.4424 :1: 0.0422 Neat Gasoline vs. 50% Evaporated Lacquer Thinner 0.5408 2: 0.0046 50% Evaporated Gasoline vs. 50% Evaporated . :1: , Lacquer Thinner 0 4414 0 0423 88 3. 3. 6 Association of Evaporated Liquids to Neat Liquids using HCA The scores plot (Figure 3.9) showed the association of the neat liquid replicates, while differentiating the liquids according to ASTM class. Using HCA, a statistical measure of the association and discrimination of the liquids in the scores plot was determined. Within the dendrogram of the neat liquids (Figure 3.8), replicates of each liquid are associated and separated from different liquids at a similarity greater than or equal to 0.959, which indicates the replicates are very similar to each other. However, replicates are expected to have a similarity closer to 1. This lower than expected similarity is due to variability in the extraction procedure of the neat liquids. Gasoline and lacquer thinner are closely positioned in the scores plot (Figure 3.5) and could not be differentiated. Using HCA, these two liquids are associated at a similarity level of 0.900 (Figure 3.9). Kerosene and gasoline are positioned firrthest apart on PCI (Figure 3.5) and, using HCA, there is no similarity between these two liquids (similarity of 0.000 in Figure 3.8). Marine fuel stabilizer and lamp fuel are positioned furthest apart on PC2 in the scores plot (Figure 3.5) and, using HCA, there is no similarity between these two liquids (similarity of 0.083 in Figure 3.9). Within the dendrogram of the neat and evaporated liquids (Figure 3.10), separation by class of the liquids is observed at a similarity level of 0.836. A high similarity between the evaporated liquids and the corresponding neat liquid is observed with the exception of 50% evaporated kerosene (0.494). As observed in the scores plot (Figure 3.8), the 50% evaporated kerosene replicates are positioned between lamp fuel and kerosene on both PCI and PC2. However, using HCA, the 50% evaporated kerosene is more similar to the neat kerosene 89 0.0 2‘ g ----------------------------------------------------- 1-------- --- 8... H d) a) ..., 0 2 £3 ,5 “s’ 532’ a l—~ 0- i- '5 3'3 § *5 E 8 g 53 9 ..1 SPEUbI'I 9199111131 Figure 3.9: HCA Dendrogram of the scores of the neat liquids. Dashed line indicates similarity of 0.959. 90 0.0 l Similarity 2]— 1% 2 10 012:1"? ...-mam. 3..) 31‘ H o ..., g T, H 0 ..., a) a) .2 =3 .2: 8 L2 “:3 E, .1: LL" f-‘ v-1 a) :L". m I— “ a 8 .a “.8 e ... c" c: g m :4 to u—l o (3.. c6 l—l SPEHbE'I 919911031 Figure 3.10: HCA Dendrogram of the scores of the neat and evaporated liquids. An asterisk indicates 50% evaporated kerosene. Dashed line indicates similarity of 0.836. 91 (similarity of 0.494) than to neat lamp fuel (similarity of 0.000). Hence, HCA can be used as a statistical measure of the association and discrimination of the neat and evaporated liquids by PCA. 3.4 Conclusions Through the use of PCA, PPMC coefficients, and HCA, the ignitable liquids are distinguished by class with the replicates of each liquid associated to each other and separated from other ignitable liquids. Furthermore, the evaporated liquids are generally associated to the corresponding neat liquid. The neat liquids were distinguished using PCA according to ASTM class, with the exception of gasoline and lacquer thinner. Through PPMC coefficients, all of the ignitable liquids were able to be distinguished from each other. Although they were positioned closely in the scores plot, gasoline and lacquer thinner were observed to only have a moderate correlation to each other with PPMC coefficients. Four of the six liquids were further distinguished from each other in the scores plot using HCA. Gasoline and lacquer thinner were not distinguishable through HCA, having a similarity greater than 0.900. The evaporated liquids were associated to the corresponding neat liquid in the PCA scores plot. Based only on the scores plot, there was no measure of the degree of association. Using PPMC coefficients, a strong correlation was determined between the evaporated liquid and the corresponding neat liquid. A similarity level was also established between evaporated ignitable liquids and the corresponding neat liquid using 92 HCA. While PPMC coefficients evaluate correlation among pairs of samples, HCA statistically measures similarity of all of the samples. Through HCA, a value can be applied to assess the similarity of the scores in the scores plot. The combination of these three statistical procedures reduces the subjectivity of associating an evaporated ignitable liquid to the corresponding neat ignitable liquid. 93 3.5 References American Society for Testing and Materials, ASTM E l618-06e1. Annual Book of AS TM Standards 14.02. Johnson KJ, Wright BW, Jarrnan KH, Synovec RE. High-speed peak matching algorithm for retention time alignment of gas chromatographic data for chemometric analysis. J Chromatogr A 2003;996:141-55. 94 CHAPTER 4 Effect of Matrix Interferences, Evaporation, and Combustion on the Identification of Mixed Ignitable Liquids in Fire Debris using Chemometric Procedures 4.1 Introduction At a suspected arson, an arsonist may have used a mixture of ignitable liquids rather than a single ignitable liquid, as in Chapter 3. Under these conditions, the ignitable liquid residues (ILR) collected from an arson scene would contain matrix interferences from the fire debris and evaporative loss from the ignitable liquid mixture. With the profile of the ILR differing from that of the ignitable liquid mixture, an arson investigator may not be able to identify the ignitable liquids present. In this research, chemometric procedures, such as principal components analysis (PCA), Pearson product moment correlation (PPMC) coefficients, and hierarchical cluster analysis (HCA), were used as an objective method to associate mixed liquid samples to the corresponding mixed liquid standards even in the presence of matrix interferences, evaporative loss, and combustion. Firstly, gasoline and kerosene were selected as the model liquids and evaporated to two levels of evaporation. Mixed liquid standards were prepared using different combinations of the neat and evaporated ignitable liquids. The mixtures were then spiked onto burned carpet in order to assess association of the mixed liquid in the presence of matrix interferences. Mixed liquids were also spiked onto unburned carpet are then burned to simulate an ILR. This was done to assess association of the mixed liquid in the presence of matrix interferences, evaporative loss, and combustion due to btu'ning. All 95 samples were analyzed by gas chromatography—mass spectrometry (GC-MS). The chromatograms were compiled into two data sets and subjected to pretreatment procedures (smoothing, retention time alignment, and normalization). Principal components analysis, PPMC coefficients, and HCA were applied to assess association of the mixed liquid samples to the corresponding mixed liquid standard even in the presence of matrix interferences, evaporative loss, and combustion. 4.2 Methods and Materials 4. 2.1 Sample Collection Gasoline and kerosene were collected from service stations in the Lansing, Michigan area. The ignitable liquids, along with the class and major compounds present in each, are listed in Table 3.1 in Chapter 3. 4. 2.2 Mixed Ignitable Liquids Gasoline and kerosene were evaporated to two different levels of evaporation (10% and 50% by volume) using filtered air and magnetic stir bars for agitation. Once evaporated, the liquids were stored in 7.4 mL amber screw cap vials (Fisher Scientific, Pittsburgh, PA) that were acid washed before use. The vials were labeled, wrapped in Parafilm® (American National Can, Menasha, WI), and stored at 16°C until analysis. A set of mixed gasoline and kerosene samples was prepared using different combinations of the neat, 10% evaporated, and 50% evaporated liquids (1:1 v/v), as shown in Table 4.1. The mixed ignitable liquids were diluted 1:50 (v/v), extracted, and 96 Table 4.1 Mixed Ignitable Liquids. Gasoline Kerosene Neat Neat Neat 10% Evaporated Neat 50% Evaporated 10% Evaporated Neat 50% Evaporated Neat 97 w".- analyzed using the same procedure described in Chapter 3. 4. 2. 3 Matrix Interferences Nylon carpet (source unknown) was cut into 5 x 5 cm2 squares, burned for 40 sec with a propane blow torch (Benzomatic, Medina, NY), and allowed to burn further for 1.0 min. The flame was extinguished through inverting a 1000 mL beaker and placing it over the burning carpet. The pieces of burned carpet were extracted using the same passive headspace extraction and eluted with CHzClz, as described previously in Chapter 3 Section 2.2. The process was repeated twice more, using fresh pieces of carpet. Additional samples of 5 x 5 cm2 carpet were burned and extinguished as described previously. A 1 uL aliquot of each mixed liquid, except for the neat gasoline: neat kerosene mixture, was spiked onto separate pieces of the burned carpet and extracted, as described previously in Chapter 3 Section 2.2. This process was repeated twice more for each mixed liquid. The chromatograms collected are subsequently referred to as the “burned then spiked” samples. A 750 11L aliquot of each mixed liquid, except for the neat gasoline: neat kerosene . . . 2 mixture, was spiked onto separate pieces of 5 x 5 cm unburned carpet. The carpet was burned following the previously described procedure and extracted following the method described in Chapter 3 Section 2.2. The chromatograms collected are subsequently referred to as the “simulated ILR” samples. 98 4. 2.4 GC—MS Analysis The experimental parameters for the GC-MS analysis are exactly the same as those described in Chapter 3 Section 2.3. 4. 2. 5 Data Pretreatment Total ion chromatograms (TICs) were generated and smoothed for the mixed ignitable liquids, the burned carpet, the burned then spiked samples, and the simulated ILR samples using the same procedure described in Chapter 3 Section 2.4. Two data sets were compiled, where both data sets contained the TICs of mixed ignitable liquid standards and the burned carpet, one data set contained the TICs of burned then spiked samples, while the other data set contained the TICs of the simulated ILR samples. Chromatograms in each data set were retention time aligned to a target chromatogram. The target chromatogram was generated by averaging the replicates of the mixed ignitable liquid standards and the burned carpet. The chromatograms were averaged due to the difficulty in forming a consensus target that included the compounds from the mixed liquids and matrix inferences. The TIC of the average target was added to each data set for alignment purposes. The two data sets were aligned using the same algorithms described in Chapter 3 Section 2.4. For each algorithm, various user-defined parameters were investigated, as shown in Table 4.2. There are typically 13 points across a peak such that window size and warp size are smaller than the peak width, while the segment size is larger than the peak width. 99 Table 4.2 User defined parameters investigated for peak matching algorithm and COW algorithm. ”2'; 1:21:12“ cow Algorithm Data Set ———.-g . . . Window Srze Warp Srze Segment Srze in Data Points in Data Points in Data Points Data Set Containing Burned then 2, 3, 4, 5, 6, 7 1, 2, 3, 4, 5 55, 65, 75 Spiked Samples Data Set Containing Simulated ILR 2, 3, 4, 5, 6, 7 1, 2, 3, 4, 5 55, 65, 75 Samples 100 ‘For the correlation optimized warping (COW) algorithm, all combinations of the warp sizes and the segment sizes in Table 4.2 were investigated. The aligned chromatograms were visually assessed to determine the parameters for optimal alignment. For the peak matching algorithm, each window size mentioned in Table 4.2 was investigated and visually assessed for misalignments. After retention time alignment, the data sets were normalized using maximum peak normalization followed by total area normalization, as described in Chapter 3 Section 2.4. 4. 2.6 Data Analysis Principal components analysis (PCA) was performed on the chromatograms of the mixed ignitable liquid standards and the burned carpet using MatLab (version 7.7.0.471, The Math Works, Natick, MA), following the procedure described in Chapter 3 Section 2.5. Scores for the data set of the burned then spiked samples were then calculated in Microsofi Excel (version 12.0.6524.5003, Microsofi Corp., Redmond, WA) and projected onto the scores plot of the mixed ignitable liquid standards and the burned carpet, following procedures described in Chapter 3. The scores plot was used to assess differentiation of the burned then spiked samples in relation to the burned carpet and to the corresponding mixed liquid standards. This procedure was then repeated to calculate and project scores for the simulated ILR samples, which were then assessed in a similar manner. Pearson product moment correlation (PPMC) coefficients were calculated for the aligned chromatograms using MatLab (The Math Works, Inc.) (Equation 2.1). The 101 . PPMC coefficients were calculated among replicates of each mixed liquid to assess the precision of the extraction and analytical procedure. Correlation coefficients were then calculated for all pair-wise combinations of chromatograms for the data set of the burned then spiked samples to assess the correlation of the samples to the mixed liquid standard. The same procedure was repeated for the simulated ILR samples, and the correlations observed were assessed in a similar manner. Using the scores calculated for each data set, HCA was performed in Pirouette (version 4.0, Infometrix, Inc., Bothwell, WA) in order to statistically evaluate the association and discrimination of the mixed liquids in the scores plot. The Euclidean distance (Equation 2.2) was first calculated among all pair-wise combinations of the PC scores. From the agglomerative method, each score was individually linked using the complete linkage method. 4.3 Results and Discussion 4. 3. 1 Optimization of Retention Time Alignment and Normalization The retention time alignment and normalization were investigated separately for the data set of the burned then spiked samples and the data set of the simulated ILR samples. For both data sets, burned carpet and the mixed liquid standards were used to determine optimal parameters. It is noteworthy that the same optimal parameters were determined for both. Window sizes of two to seven points were investigated for the peak matching algorithm (Table 4.2). Since major misalignments were observed for each window size, 102 the peak matching algorithm was excluded from further investigation. Using the COW algorithm, major misalignments were observed for most combinations of warp sizes and segment sizes. Visual assessment of the chromatograms yielded a warp size of two points and a segment size of 75 points as the optimal alignment (Figure 4.1A and B). Hence, these parameters were used as the optimal alignment for both data sets. After alignment, the chromatograms were normalized firstly with maximum peak normalization and then with total area normalization of the replicates to minimize differences in abundances (Figure 4.2A and B), as discussed in Chapter 3 Section 3.2. 103 “1‘9- I . 4 E06 A___- -.. - _ -2- __-- - ,. -- - - , 2-2.... ...-.. -. ...-....-M,......_..-._n m, - 7N.-. 8 . a . (U ; 'O i ‘5 .0 <2 -‘ 3.6 Retention Time (min) 3.8 1.4E06 :B 8 . c: , «3 “U 2 5 i .0 <1 0 __....w __. . . i -.-..-n- 4 3.6 Retention Time (min) 3.8 Figure 4.1: A) Unaligned toluene peak in the TIC of mixed liquids. B) Well aligned toluene peak in the TIC of mixed liquids with a warp size of two points and a segment size of 75 points. 104 9.0505 A Abundanc e!) . Abundance l4.2 Retention Time (min) 14.4 Figure 4.2: A) Unnormalized C14 normal alkane peak. B) Normalized C14 normal alkane peak. Neat Gasoline: Neat Kerosene Mixture (—), Neat Gasoline: 10% Evaporated Kerosene Mixture (— - ), Neat Gasoline: 50% Evaporated Kerosene Mixture (- a a ). 105 4. 3. 2 Matrix Interferences The TICs of burned carpet show matrix interferences such as 2,4-dimethyl-1- heptene, styrene, benzaldehyde, and acetophenone [1,2] (Figure 4.3A). These matrix interferences present are from the thermal degradation of the adhesive, the yarn, and the backing material in carpet [2]. The TICs of the mixed liquid standards (Appendix E, Figures E] to E5) consist of the major compounds present in both gasoline and kerosene: toluene, the C2- and C3-alkylbenzenes, naphthalenes, and the C11-C17 normal alkanes (Figure 4.38). Of the compounds that are common between chromatograms of the burned carpet and mixed liquid standards, only toluene is present in both. Thus, the use of toluene as an identifying peak for the gasoline in the mixed liquid is diminished. From the burned then spiked samples, the effect of both matrix interferences and evaporation without combustion of the mixed liquid can be observed in the chromatograms (Figure 4.4A). The burned then spiked samples contain some of the more dominant compounds in the mixed liquids, such as toluene, the C3-alkylbenzenes, cyclic alkanes, naphthalenes, and C11-C13 normal alkanes. The presence of these compounds is expected, since the mixed liquids have been spiked onto burned carpet and have not been subjected to the burning procedure. However, matrix interferences have complicated the identification of these compounds through visual assessment of the chromatogram. In the chromatograms, styrene (6.165 min) from the burned carpet coelutes with p-xylene (6.205 min) from the mixed liquid. Also, benzaldehyde (7.544 min) from the burned carpet coelutes with the m-ethyltoluene (7.556 min) in the mixed liquid. Even though 106 1.01306 A 3 81 g j C.5 Branched Alkenes 5 3 r1 .0 .- < 2 4 5 6 j 1 I! 3.0 Retention Time (min) 21 .0 1.0E06 B ; Cz-Alkylbenzenes ; [__l ' Naphthalenes and Cyclic Alkanes I 1 I C3-Alkylbenzenes g C12 C13 E C11 C14 < ; C15 Cm i C17 3.0 Retention Time (min) 21.0 Figure 4.3: A) TIC of burned carpet. B) TIC of 10% evaporated gasoline: neat kerosene mixture. Major components are labeled: 1) toluene, 2) 2,4-dimethyl-l-heptene, 3) styrene, 4) benzaldehyde, 5) acetophenone, 6) 1,3-diphenylpropane 107 1.0E06 w A 1 , 3 g Naphthalenes and Cyclic Alkanes ‘ I I 1 g l Cz-flaylbenzenes C.5 Branched Alkenes < C3-Alkylbenzenes T Cn Clzc 3.0 Retention Time (min) 21 .0 101306 B , 3 Naphthalenes and Cyclic Alkanes 8 ' ' .5 3 Cl, Branched Alkenes s ” g: C13 C14 ; Cls S i 2 4 6 C” 1 0 L.l_L_.JJ.J i . ....M _ _ _M _____ . l 3.0 Retention Time (min) 21 .0 Figure 4.4: A) TIC of burned carpet spiked with 10% evaporated gasoline: neat kerosene mixture. B) TIC of 10% evaporated gasoline: neat kerosene simulated ILR sample. Major components are labeled: 1) toluene, 2) 2,4-dimethyl-1-heptene, 3) styrene, 4) benzaldehyde, 5) 1,3-diphenylpropane, 6) acetophenone. 108 some of the compounds from the burned carpet have eluted with some of the compounds in the mixed liquid, the relative peak heights of the C2- and C3-alkylbenzenes in the burned then spiked samples have not changed when compared to the mixed liquid standard. The relative peak heights for the normal alkanes in the burned then spiked sample when compared to the mixed liquid have decreased in abundance. Consequently, many of the later eluting normal alkanes are at a low abundance and not able to be identified in the chromatogram. In addition, evaporation of the mixed liquid also complicates the subsequent identification. Since at least one of the liquids in each mixed liquid has been evaporated, some of the peaks in the burned then spiked samples have been lost due to evaporation. Although this is not highly problematic for burned carpet spiked with 10% evaporated gasoline: neat kerosene sample, evaporation complicates the burned carpet spiked with neat gasoline: 50% evaporated kerosene sample The evaporated kerosene part of this burned then spiked sample has lost the early eluting alkanes. Thus, evaporation has increased the difficulty in identifying the mixed ignitable liquid as containing kerosene. The difficulty in visual comparison of the simulated ILR samples (Figure 4.48) to the mixed liquid standard has increased from that of the burned then spiked samples. Most of the compounds present at early retention times, such as the C2- and C3- alkylbenzenes, have been lost due to evaporation during the burning process. The C12- C15 normal alkanes along with cyclic alkanes and naphthalenes have been concentrated by evaporation during burning, but are still difficult to identify in the simulated ILR samples. The peak heights of the normal alkanes have also changed due to burning when 109 compared to the mixed liquid standard. As observed in the chromatograms, the matrix interferences from the burned carpet coelute with or mask some of the compounds from the mixed liquids. The identification of the ignitable liquids from fire debris has become more difficult, making visual assessment even more complicated. Thus, an objective method is necessary to overcome the subjectivity introduced through visual assessment. 4.3.3 Association and Discrimination of Mixed Liquids in Presence of Matrix Interferences and Evaporation Principal components analysis was used to develop an objective method. The scores plot of the first principal component (PCI) and the second principal component (PC2) accounts for 93.6% of the variance among the burned carpet and the mixed liquid standards (Figure 4.5). Very little spread is observed among the replicates of each of the mixed liquid standards, and each standard can be differentiated from each other. However, spread is observed among the burned carpet replicates. The PPMC coefficients for replicates for the mixed liquid standards are greater than 0.99 indicating a precise analytical procedure (Table 4.3) (Appendix F, Table F.1). Although the average PPMC coefficient among the burned carpet replicates (0.9321 d: 0.0480) indicates strong correlation, a coefficient close to 1.0000 is expected for replicates. The greater spread in the replicates of the burned carpet is most likely due to the irreproducibility of burning as well as some variability in the extraction procedure. The positioning of the burned carpet and mixed liquid standards on the scores plot can be explained with reference to the corresponding loadings plots (Figure 4.6). 110 2.0E06 g I. ‘12 E", o ’* N i I B A ~2.0E06 -4.0E06 0 4.0E06 PCl (75.1%) Figure 4.5: Scores plot of PC] vs. PC2 based on the TIC for the burned carpet and the mixed liquid standards. Liquids were indicated by symbol. Burned Carpet (I), Neat Gasoline: Neat Kerosene Mixture(0), Neat Gasoline: 10% Evaporated Kerosene Mixture (t), Neat Gasoline: 50% Evaporated Kerosene Mixture (A), 10% Evaporated Gasoline: Neat Kerosene Mixture (O), and 50% Evaporated Gasoline: Neat Kerosene (O). 111 Table 4.3: PPMC coefficients for replicates (n=3) for the mixed liquid standards and burned carpet based on the TIC. Mean PPMC Coefficient :1: Standards Standard Deviation for Replicates Neat Gasoline: Neat Kerosene 0.9969 i 0.0022 Neat Gasoline: 10% Evaporated Kerosene 0.9972 i 0.0022 Neat Gasoline: 50% Evaporated Kerosene 0.9994 3: 0.0003 10% Evaporated Kerosene: Neat Kerosene 0.9916 at 0.0020 50% Evaporated Gasoline: Neat Kerosene 0.9944 5: 0.0025 Burned Carpet 0.9321 i 0.0480 112 u. 0.25 . A l 4 Naphthalenes and Cyclic Alkanes l l ' C13 C3-Alkylbenzenes C12 C14 "‘ I Cu C15 33 3 5 § .. C17 on 0 * l . t: ‘5 7 8 8 2 9 '4 [_J C15 Branched Alkenes 6 -0.25 3 Retention Time (min) 21 0.18 B C11 C3-Alky1benzenes l—_1 4 8 | a. 5 c: i O 22” 0 r ‘3' 1 2 C 3 n 6 C16 C13 C .5 1 C14 1 _0 18 Naphthalenes and Cyclic Alkanes . . 3 Retention Time (min) 2 I Figure 4.6: Loadings plots of (A) PCI and (B) PC2 based on the TIC for burned carpet ad mixed liquid standards. Major components were labeled: 1) toluene, 2) 2,4-dimethyl-l- heptene 3) ethylbenzene, 4) o-xylene, 5) p-xylene, 6) styrene, 7) a-methylstyrene, 8) acetophenone, 9) 1,3-diphenylpropane. 113 The negative positioning of the burned carpet on PC] is due to 2,4-dimethyl-l- heptene, styrene, benzaldehyde, acetophenone, and 1,3-diphenylpropane, which load negatively. The negative positioning of the burned carpet on PC2 is due to 2,4-dimethyl- l-heptene and styrene, which load negatively. The mixed liquids are positioned positively on PC] due to toluene, the C2- and C3-alkylbenzenes, naphthalenes and C1 l-C17 normal alkanes. On PC2, the neat gasoline: 50% evaporated kerosene mixture is positioned negatively while the other mixed liquids are positioned positively. The neat gasoline: 50% evaporated kerosene mixture is positioned negatively due to the negative loading of the C13-C17 normal alkanes and naphthalenes, which are higher in abundance in this liquid. The other mixed liquids are positioned positively on PC2 due to the positively loading C2- and C3-alkylbenzenes, which are at a higher abundance in these mixed liquids. When the burned then spiked samples are projected onto the scores plot, they are positioned negatively on PCI between the burned carpet and the mixed liquid standards (Figure 4.7). The replicates of the burned then spiked samples are spread due to the irreproducibility of the burning process. For replicates of a sample, the PPMC coefficients are ideally expected to be 1.0000. Low correlation coefficients among replicates are observed for all of the burned then spiked samples, especially with the burned carpet spiked with 50% evaporated gasoline: neat kerosene mixture (0.8713 :t 0.1000) (Appendix F, Table F.2). Although a strong correlation is observed, the correlation coefficients are lower than expected for replicates of a sample. 114 2.0E06 o i» PC2 (18.4%) -2.0E06 -4.0E06 0 4.0E06 PC] (75.1%) Figure 4.7: Scores plot of PCI vs. PC2 based on the TIC for the burned carpet and mixed liquid standards. Liquids were indicated by symbol. Burned Carpet (I), Neat Gasoline: Neat Kerosene Mixture (O), Neat Gasoline: 10% Evaporated Kerosene Mixture (tr), Neat Gasoline: 50% Evaporated Kerosene Mixture (A), 10% Evaporated Gasoline: Neat Kerosene Mixture (O), and 50% Evaporated Gasoline: Neat Kerosene (O). The half fill indicates burned then spiked samples. 115 1‘." The burned carpet spiked with the neat gasoline: 50% evaporated kerosene mixture is positioned negatively on PCI and slightly negatively on PC2 (Figure 4.7). When the positioning is visually compared to burned carpet and the corresponding mixed liquid standard, the burned then spiked sample is more closely positioned to the burned carpet on PC] and PC2. When chromatograms of the burned carpet spiked with the neat gasoline: 50% evaporated kerosene mixture are assessed, the matrix interferences from the burned carpet are more abundant than the C2- and C3-alkylbenzenes and the C13-C17 normal alkanes from the mixed liquid. Because the matrix interferences present, such as 2,4-dimethyl-1-heptene, styrene, and acetophenone, are varying more than compounds from the mixed liquid, the burned then spiked sample is positioned negatively on PC]. Along with the matrix interferences, toluene and C13-C17 normal alkanes from the mixed liquid are loading negatively on PC2 and have positioned the liquid slightly negatively in the scores plot on that PC. Although the C2- and C3-alkylbenzenes are present in the liquid, they are less abundant than the matrix interferences and the C13-C17 normal alkanes, limiting their contribution to the positioning of the samples in the scores plot. For PPMC coefficients, a moderate correlation is observed between the burned then spiked sample and the burned carpet (0.7515 i 0.0478) (Table 4.4) (Appendix F, Tables F3 to F7). A moderate correlation is also observed between the burned then spiked sample and the corresponding mixed liquid standard (0.6462 :t 0.0162). Since moderate correlations are observed, the burned then spiked samples cannot be associated to the mixed liquids only using PPMC coefficients. 116 Table 4.4: PPMC coefficients between burned then spiked samples and corresponding mixed liquid standard (n=9) and burned then spiked samples and burned carpet (n=9) based on the TIC. Mean PPMC , Coefficient :1: Mean PPMC Standard Coefficient :1: . Deviation Standard Burned then Spiked Samples Compared to Deviation Corresponding Compared to Mixed Liquid Burned Carpet Standard Neat Gasoline: 10%Evaporated Kerosene Neat Gasoline: 50% Evaporated Kerosene 10% Evaporated Gasoline: Neat Kerosene 50% Evaporated Gasoline: Neat Kerosene 117 0.6869 :t 0.1228 0.6462 :1: 0.0162 0.7506 d: 0.0985 0.7879 :1: 0.1509 0.7038 i 0.1107 0.7515 :1: 0.0478 0.6532 :t 0.1381 0.5520 :t 0.1827 When each burned then spiked sample is compared to the burned carpet and the corresponding mixed liquid standard using PPMC coefficients, moderate correlations are observed (Table 4.4). With HCA, the burned then spiked samples showed a higher similarity to the btu'ned carpet (0.450) than their corresponding mixed liquid standards (Figure 4.8). Through all of these data analysis procedures, the burned then spiked samples are differentiated from the burned carpet and the corresponding mixed liquid standard. As observed in the PCA scores plot, the burned then spiked samples are positioned between the burned carpet and the mixed liquid standards. The moderate PPMC coefficients indicate that the burned then spiked samples cannot be associated to either the burned carpet or the corresponding mixed liquid standard. Although the HCA dendrogram gives a higher similarity of the burned then spiked samples to the burned carpet than to the corresponding mixed liquid standard, the similarity observed does not confirm that the burned then spiked samples are positioned more closely to the burned carpet. 4. 3.4 Association and Discrimination of Simulated ILR samples in the Presence of Matrix Interferences, Evaporation, and Combustion The simulated ILR samples were used to assess the effects of matrix interferences and evaporative loss, a 5 well as combustion due to burning. In order to assess the association and discrimination of the simulated ILR samples to the burned carpet and the mixed liquid standards, the simulated ILR samples were projected onto the scores plot (Figure 4.9). 118 “at“?! 0.0 E E m o I [“1 ”L1 FIE—1 :LI [‘1 ..z 2 Enqlwonu m0 g z .. ‘2 a S '3‘ § m 3 ‘13 is" 3 9; 2 g 2 $2 3’. z z soldures paxtds uoqi paumg Sunnmuog 19$ meg Figure 4.8: HCA Dendrogram of the scores of the data set containing the replicates (A, B, C) of the burned then spiked samples. The N is the neat part of the sample, and kero represents kerosene. The italicized labels indicate the burned then spiked samples. 119 2.0E06 b gs '5 v, 3 0 * 8 al.?) 9. A A 0 A0 0 A -2.0E06 -4.1E06 0 4.1E06 PCl (75.1%) Figure 4.9: Scores plot of PC] vs. PC2 based on the TIC for the burned carpet and the mixed liquid standards. Liquids were indicated by symbol. Burned Carpet (I), Neat Gasoline: Neat Kerosene Mixture(0), Neat Gasoline: 10% Evaporated Kerosene Mixture (*), Neat Gasoline: 50% Evaporated Kerosene Mixture (A), 10% Evaporated Gasoline: Neat Kerosene Mixture (I), and 50% Evaporated Gasoline: Neat Kerosene (O). No fill indicates simulated ILR samples. 120 The simulated ILR samples, which are positioned negatively on both PC1 and PC2, are separated into two groups. One group containing most of the simulated ILR samples is positioned closely to the burned carpet. The other group, which contains the rest of the simulated ILR samples, is positioned between the burned carpet and the neat gasoline: 50% evaporated kerosene mixture. The spread observed among replicates is due to the irreproducibility of burning. With replicates, PPMC coefficients of 1.0000 are ideally expected. Lower PPMC coefficients for replicates are observed for the simulated ILR samples, especially neat gasoline: 50% evaporated kerosene simulated ILR (0.6450 3: 0.2748) (Appendix G, Table G.1). Moderate to strong correlations of the simulated ILR samples are lower than expected for replicates. The simulated ILR samples for 50% evaporated gasoline: neat kerosene mixture, neat gasoline: 10% evaporated kerosene mixture, and neat gasoline: 50% evaporated kerosene are associated with the burned carpet and differentiated from their corresponding mixed liquids in the scores plot. The other simulated ILR samples of 10% evaporated gasoline: neat kerosene mixture and one of the replicates of the neat gasoline: 50% evaporated kerosene mixture are differentiated from the burned carpet and the mixed liquid standards. Most of the compounds from mixed liquids have been lost due to burning, and matrix interferences are at a greater abundance. All of the simulated ILR samples are negatively positioned on PC1 due to the matrix interferences from the burned carpet, such as 2,4-dimethyl-1-heptene and styrene. On PC2, the simulated ILR samples are separated due to the variability of burning. The simulated ILR samples of the 50% evaporated gasoline: neat kerosene mixture, neat 121 gasoline: 10% evaporated kerosene mixture, and neat gasoline: 50% evaporated kerosene are positioned slightly negatively on PC2. Their positioning is due to the matrix interferences from the burned carpet contributing to the positioning on that PC. The compounds from the mixed liquid are present in very low abundances and are not contributing the positioning of these simulated ILR samples. The simulated ILR samples of 10% evaporated gasoline: neat kerosene mixture and one replicate of the neat gasoline: 50% evaporated kerosene mixture are positioned negatively on PC2. Their negative positioning is due to the matrix interferences from the burned carpet along with the C13- C17 normal alkanes and the naphthalenes from the mixed liquids. All of these compounds are loading negatively on PC2. When PPMC coefficients were calculated, a moderate to weak correlation was observed between the simulated ILR samples and the corresponding mixed liquid (Table 4.5) (Appendix G, G2 to G6). A moderate to strong correlation was observed between the simulated ILR samples and the burned carpet. When the HCA dendrogram was assessed, the simulated ILR samples for the 50% evaporated gasoline: neat kerosene mixture, the neat gasoline: 10% evaporated kerosene mixture, and the neat gasoline: 50% evaporated kerosene liquids were grouped with the burned carpet at a similarity of 0.733 (Figure 4.10). The other simulated ILR samples are grouped with the burned carpet at a similarity of 0.503. All of the simulated ILR samples show a similarity of 0.000 to the corresponding mixed liquids. Most simulated ILR samples are associated with the burned carpet and differentiated from the corresponding mixed liquid standard. This is confirmed in the PCA scores plot. The PPMC coefficients for the simulated ILR samples are moderate to . 122 ’4. E"! - Table 4.5: PPMC coefficients between the simulated ILR sample and corresponding mixed liquid standard (n=9) and the simulated ILR sample and burned carpet (n=9) based on the TIC. Simulated ILR Sample Neat Gasoline: 10% Evaporated Kerosene Neat Gasoline: 50% Evaporated Kerosene 10% Evaporated Gasoline: Neat Kerosene 50% Evaporated Gasoline: Neat Kerosene Mean PPMC Coefficient :1: Standard Deviation Compared to Corresponding Mixed Liquid Standard 0.3556 i 0.1319 0.4068 :t 0.3016 0.5365 :L- 0.0588 0.3502 d: 0.0496 Mean PPMC Coefficient :1: Standard Deviation Compared to Burned Carpet 0.8485 3: 0.1107 0.7771 i 0.2203 0.7153 i 0.0780 0.8538 i 0.0696 “ 0.0 ’5 .5. m L L— o Hafl 2.21:1 .2... VM an mam mo