Organic Mass Spectrometry, 1970, Vol. 4, pp. 493 to 501. Heyden & Son Limited. Printed in Northern Ireland

APPLICATIONS OF ARTIFICIAL INTELLIGENCE
FOR CHEMICAL INFERENCE—V*

AN APPROACH TO THE COMPUTER GENERATION OF CYCLIC
STRUCTURES. DIFFERENTIATION BETWEEN ALL THE POSSIBLE
ISOMERIC KETONES OF COMPOSITION C,H,,0.

Younus M. SHEIKH, ARMAND Bucus,t ALLAN B. DELFINO,t GUSTAV SCHROLL,§
A. M. DurrieLp, CARL Dyerassi, B. G. BUCHANAN, G. L. SUTHERLAND,

E. A. FEIGENBAUM and J. LEDERBERG
Departments of Chemistry, Computer Science and Genetics, Stanford University,
Stanford, California 94305, USA

(Received | July 1970; accepted (revised) 7 August 1970)

Abstract—The computer program DENDRAL has been modified so as to include cyclic structures
for the first time. Asa result a list of all the possible isomers (linear and cyclic) of selected compositions
can now be generated. The number of cyclic structures exceeds that of the linear molecules for a
given composition. A method, based on their physical properties (i.r., n.m.r. and mass spectra), for
the identification of each of the 27 possible ketones (exclusive of 5 cyclopropanones) of composition
C.H,,O is described.

RECENT publications’? from our interdisciplinary research group describe the
development and refinement of a computer program (Heuristic DENDRAL) capable
of the structural elucidation of unknown saturated aliphatic ketones,” ethers? and
amines! from their low resolution mass spectra. While we readily admit that the
chosen functional groups and their molecular environments represent only a small
area of interest to an organic chemist, nonetheless these papers represent a start at
direct computer interpretation of physical data without human intervention. The
physical data used to date have primarily been low resolution mass spectra, and when
available, n.m.r. spectra. Our experience is that the correct structure always appeared
in the final output. This output list may consist of only one or a few selected out of
many thousands, often millions, of possible structures.

The present report is a significant extension of the domain of candidate structures,
namely cyclic molecules. The variety and complexity of ring-containing structures of
course gives them a major place in the work of the organic chemist. They also
introduce many problems of symmetry beyond those seen in acyclics, and indeed the
efficient generation, without everwhelming redundancy, of exhaustive lists of isomeric
structures is a major challenge to the Heuristic DENDRAL STRUCTURE GEN-
ERATOR, || the program module responsible for fetching (in principle) all possible
candidate molecules as potential solutions to an analytical problem.

A completely general approach to the design of a ‘ring-generator’ has been
formulated in some detail.* It is based on a major classification of rings according to

* For Part IV, see Ref. 1.

+ On leave of absence from the University of Geneva, Geneva, Switzerland.

t Present address: Allen-Babcock Computing, Palo Alto, California, USA.

§ Recipient of a Fulbright Travel Award, Present Address: Chemical Laboratory II, University
of Copenhagen, Denmark.

|| Program modules are written in upper case,

493
494 Y. M. SHEIKH et ai.

the paths between the vertex atoms, namely those atoms at which rings are ‘fused’ in
conventional descriptions. This corresponds to the set of trivalent, cyclic graphs, the
enumeration of which has been programmed, computed and described.5 Writing the
complete program for all possible ring molecules should offer no special difficulties
except tedium and running time and we have deferred doing this in favor of more
general problems that might encompass this almost incidentally. (For example, we
should write programs that ‘understand’ group theory, and can save us the tedium of
transmitting an infinity of small tricks and insights for weeding out symmetries
prospectively, as indeed was done in producing DENDRAL up to this point.)

The perfectly general case may also have limited utility compared to applications
where some constraints on the character of possible rings are implicit in a problem
statement.

At this time, we deal with a specific problem (ketonic isomers of CgH,,O) which
encompasses, at most, a single, simple ring. We have then described all the relevant
rings through a special purpose program. In effect, this takes all possible pairwise
selections of 2 hydrogen atoms from acyclic isomers of C,H,,O (2H + C,H,,.0), and
replaces the pair with a cross-link, in effect closing a ring. Some graph theoretical
tricks are introduced to minimize redundancies, and retrospective checks for iso-
morphism were also inserted to eliminate them. This approach generates the entire
molecule, including side chains. The potential redundancy approaches the number of
ways that the ring-molecule can be cut to produce different trees.

A second approach, closer to the spirit of generalized DENDRAL, regards every
pure ring (i.e. no side chains) as a SUPERATOM. A full list of these superatoms is
generated by the function MONOCYCLIC. Each SUPERATOM (RING) is then
embellished with the appropriate permutations of side chains, account being taken of
the symmetry of the RING.

Although we have described® the computer program, DENDRAL, which gen-
erated complete and irredundant lists of the number of acyclic isomers which can be
formulated from a given composition, no computer routine then existed for the
enumeration of cyclic molecules. In view of the substantial increase in the number of
isomers (and thus the cost of computer time) of a selected composition when cyclic
representations are generated, it has only been feasible to produce the numbers of
possible isomers for comparatively small empirical formulae (Tables 1 and 2). For
example it takes approximately 6-2 seconds to generate all the non cyclic isomers of
C,H,O and 20:3 seconds for the non-cyclic isomers of C,H,O; but it takes approx-
imately 14-7 seconds for all the cyclic isomers of CjH,O and 66-0 seconds for all the
cyclic isomers of C,H,O. Work with larger compositions will have to wait until an

TABLE 1. NON-CYCLIC STRUCTURES OF COMPOSITION C,H, .O

 

 

n Ketones Aldehydes Ketenes Alcohols Ethers Total
2 0 0 1 0 0 1
3 0 1 1 1 1 4
4 1 3 2 4 5 15
5 4 8 3 14 17 46
6 13 21 7 47 62 150
7 40 56 13 182 207 498

 
Applications of artificial intelligence for chemical inference—V 495

TABLE 2. CYCLIC STRUCTURES OF COMPOSITION C,H.,-20

 

 

n ~~ Ketones Aldehydes Alcohols Ethers Total Acyclic + Cyclic
2 0 0 0 0 0 1
3 1 0 2 4 7 ll
4 2 i 10 20 33 48
5 7 4 35 79 145 191
6 19 13 215 310 558 708
7 57 47 22? 72? 722? 2222

 

alternative algorithm is developed. The following examples with their familiar
chemical representations are presented as illustrations of the representation of cyclic
structures in DENDRAL dot notation.®

The program derives its canonical notation from the following three points:

1. A structure is built from the lowest valued DENDRAL acyclic form.

2. The skeletal atoms are numbered in the order of their appearance in the dot
notation.

3. Ring construction is achieved by using the smallest valued pair of numbers to
describe a given cross-link.

0
1 2,34 5 6 1 2 3 4 1 6
CH;.. CsH, C.=CH, O CH,—CH,-—-CH;—_CH, — 5 ;
(4,6) 6 5| (4,6) °
CH;—C==-O 3
6
1 2,3,4 5 6 1 2 3 4 CHO
CHa... CsH, CH:.CH=0O CH,—CH,—C H,—CH, 5
| 5 6 Ss ! 4
CH,—CHO 5 3
(4,5) (4,5) °

Three modes exist for the routine operation of the DENDRAL computer program.
First it can construct all the topologically possible acyclic isomers; second, using
the option of a BADLIST* filter, the chemically stable isomers can be selected; or
third, using the GOODLIST constraints, only those isomers containing a specific
structural unit (e.g. ketones, alcohols, etc.) will appear. The concept of a BADLIST
for ring DENDRAL is more complex than for linear structures. In the latter instances
organic chemists will readily agree on those groups of atoms considered to be chemi-
cally unstable; however, when these groups of atoms are embedded in ring structures
they often will be stable. Furthermore, steric considerations must be considered in
the implementation of ring DENDRAL. For instance the following restrictions have
been placed on BADLIST: no isomers are accepted with a double bond in a three-
membered ring (since these compounds would only be of limited stability); no triple

* BADLIST, described in previous publications,?-*.6 is the list of substructures which must not
become embedded in molecular structures because, for example, they result in unstable molecules.
Similarly, GOODLIST is the list of substructures which the program must build into every molecule it
generates.
496 Y. M. SHEIKH et al,

bond can be inserted in less than an eight-membered ring; and allenes are prohibited
unless the ring size exceeds 8 atoms.

Table | is a new listing of the number of acyclic isomers of the formula C,H2,-,0
as generated by the computer program DENDRAL. The number of ketones, al-
dehydes, ketenes, alcohols and ethers of a given composition were generated by
placing the appropriate subgraph on GOODLIST.

Table 2 contains the number of possible cyclic structures of the same composition
C,H2n2O classified according to the number of ketones, aldehydes, alcohols and
ethers. It is apparent that once ring systems are considered, the number of topologi-
cally possible isomers of these groups for any composition exceeds the number of their
linear counterparts. Furthermore, while for CsH,O the number of cyclic ketones and
cyclic aldehydes remains relatively low (19 and 13 respectively) the number of cyclic
alcohols (215) and ethers (310) constitutes a large majority of the total structures.

It seemed desirable at this stage to confront Heuristic DENDRAL (the computer
program) with the task of differentiating between each isomer of a given composition
by the physical data (mass, n.m.r., ir. and u.v. spectrometry) available for each
isomer. In view of the scope of the synthetic problem (it being realized that many,
probably the majority, of the isomers listed are as yet unknown and would have to be
synthesized) we concentrated our attention on the question of differentiating between
each of 27 possible ketones of composition C,H,,O. It was decided to omit the
five cyclopropanones (XXVIII to XXXII) as these compounds would be of limited
stability.

CH, CH, C,H, Neon. a-C3sH, — iso-C4H,
a S=0 A A.
oO oO *
CH, CH, ° °
(XX VHD (X XIX) (XXX) (XXX1) (XXXH)

The location of the carbonyl absorption in the infrared spectra (Table 3) of the 27
isomeric ketones (I to XXVII) serves as a method of distributing them into the five
categories shown in Table 3.* All the 8 linear «,f-unsaturated ketones absorb in the
range 1675 to 1685 cm.-t and mass spectrometry can be used to differentiate further
within this group. Compounds IV, V, VII and VIII can be identified from their mass
spectra as methyl ketones (large [M — 15] ion) and the n-propyl isomer alone of this
sub-set affords a McLafferty rearrangement ion at mass 70. The two ethyl ketones X
and XI (abundant {M — 29] ions) and the two vinyl ketones XII and XTII (low
intensity but identifiable [M — 27] ions) can be recognized from their mass spectra,
with the latter pair being separable since XIL alone of the two will yield a McLafferty
rearrangement ion. A unique identification between the remaining five isomers
(Table 3, column 3) can be accomplished by n.m.r. spectroscopy.

The other four major categories in Table 3 are delineated into their respective
groups by the position of the carbonyl absorption in the infrared spectrum of each

* Although we chose infrared absorption spectra to distinguish the five categories shown in
Table 3, ultraviolet absorption spectra could have been utilized for an initial distribution of the
isomers into sud-classes.
Applications of artificial intelligence for chemical inference—V 497

TABLE 3. DIFFERENTIATION BETWEEN ALL KETONE ISOMERS OF C,H,,O USING I.R., MASS AND N.M.R

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SPECTROSCOPY
COMPOUND MS NMR
C—CO—C=-C—C—C (TV) 1 C methyl
Cc
C—CO—-C=—C (V) | All methyl ketones 2 identical C methyls
\
Cc
Cc
C—CO—C==C—C (VID | No McLafferty Rear- 2 different C methyls
- rangement
5
” C—CO—C—C—C (VII) McLafferty at m/e 70
ie) |
2 Cc
Ww
6 | C-C—CoO-C=C—C (X) 1 C methyl = doublet
7 1 C methyl = triplet
&
C—C—CO—C=C (XD Ethy! ketones 1 C methyl = singlet
| 1 C methyl = triplet
Cc
C—C—CO—C=—C Vinyl ketone
(XID | No McLafferty Rear-
Cc rangement
C=C—CO—C—-C—C (XH) | Vinyl ketone
McLafferty at m/e 70
C—CO—C—C--C=C (I) no C methyl!
C—CO—C—C—C—C dD 1 C methyl = doublet
C—CO—C—-C=C (VD All methyl no —CH,—
| ketones
c No McLafierty
e Rearrangement
a . .
o no vinylic proton
R XAT
©
S C—C—CO—C—C==C (xX) Ethyl ketone
~
| C—Co—C—Cc—C (I) | Methyl ketone
“ | McLafferty at m/e 58
Cc
(IV) | Cyclic ketone from m.s.

 

(=o

 

 

 

 

 
498

TABLE 3 (Contd.)

Y. M. SHEIKH et al.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Compound MS NMR |
PL (XVID +D,>M+4
oO
7
{ | (XVID +D,0 -+M +3
0
7
5 oO
2 —
& [1 (XIX) 2 different —CH,—
2 c
o
5 O
a tJ (XX) 2 identical —CH,—
==O . .
(XX) 2identical C-methyl groups
Oo 2 different C-methyl groups
ctl (XXII) yl group
L\. co—c (XXIV) 1 C methyl as a singlet
7
& All
oO
SZ L\ co_c (XXV) methyl! 1 C methyl as a doublet
< ketones
£
8) Lo conc (KXVID No C methyl signal
? .
~ Ethyl ketone
L\ co cc XW y
B (xv) +D,0 >M +3
Q
< O
2
wy
& (XVI) 4D,0 ->M +4
4
0

 

 

 

All NMR tests are simple. For instance one C-methyl means a methyl group which is not on a

carbonyl! function.

 
Applications of artificial intelligence for chemical inference—V 499

isomeric ketone. Further distinction within each of these categories is obtained
(Table 3) from either or both the mass and n.m.r. spectral data. In the case of 2- and
3-ethylcyclopentanones (XV and XVI) and again with 3- and 2-ethylcyclobutanone
(XVII and XVIII) it would be necessary to rerun the low resolution mass spectrum of
each compound following basic exchange in deuterium oxide. Under these conditions
the molecular ion of XV and XVIII would increase by 3 mass units and XVI and
XVII would increase by 4 mass units.

It is evident from this discussion that the problem of distinguishing between all the
isomeric ketones of composition C,H;gO would be achieved with a computer by
programming a specific set of heuristics based on the data summarized in Table 3.
This approach is clearly feasible for the 27 ketone isomers of Table 3 where one is
concerned with a limited and well defined search space. In those instances (e.g.
alcohols (215 cyclic isomers) or ethers (310 cyclic isomers)) where many more com-
pounds of composition C,H,O exist, more general heuristics would have to be
programmed. This is not as yet feasible since we have very little or no information on
the spectroscopy and especially mass spectrometric behavior of the many diversified
structures involved. In these instances the problems of first synthesizing many of the
yet undescribed compounds, some of which could present a serious synthetic challenge,
and the collection of their physical data must be surmounted before extensive pro-
gramming could be commenced. The magnitude of these problems mitigated against
our pursuing this investigation further.

The ability of the computer to generate the individual isomers whose total number
are listed in Tables 1 and 2 demonstrates to the organic chemist that many structures
of a given composition remain to be synthesized.

Clearly at this stage of development the computer is unsurpassed in the task of
generating all, or selected, structures of a given composition. This in itself represents
convincing testimony of the power computers will exercise in the future development
of organic chemistry.* One must conclude, however, that at the moment computers
are most useful for routine structure determination from physical data in the same
specific areas where they already have demonstrated! their superiority to the organic
chemist.

Before the computer program can successfully solve structure determination
problems it is necessary for it to have rules of thumb (heuristics) for differentiating
classes of isomers. These rules are currently generated by an organic chemist looking
at the mass spectra of representative compounds of the class.t As a matter of interest
for this publication all isomers of Cg,H,gO were synthesized and their mass spectra
studied, although it would seldom be necessary to have the complete list of spectra for
the class of compounds under consideration.

EXPERIMENTAL

Mass spectra were obtained by Mr R. G. Ross with an MS-9 mass spectrometer operating at 70 eV
and an ion source temperature of 180°. Samples were introduced by the heated inlet system.
All compounds were purified by preparative gas chromatography over a 20% carbowax column

* See for instance E. J. Correy and W. T. Wipke, Science 166, 178 (1969), where a suitably
programmed computer has been used to assist in the design of complex organic syntheses.

+ The importance of computer learning routines for just this process cannot be underestimated.
See for instance P. C. Jurs, B. R. Kowalski, T. L. Isenhour and C. N. Reilley, Anal. Chem, 41, 1949
(1969).
500 Y. M. SHEIKH ef al,

(Zin. x 10 ft.). Compounds I, V, VI, XIV, XV, XVI and XXIH were commercially available.
Compounds IV, X and XHI were prepared by manganese dioxide oxidation of the corresponding
commercially available alcohols. Physical data for compounds XII,” XIX,° XX,’ XXI,’ XKH? and
XXV*!0 were acquired from the literature.

4-Hexen-2-one (II)" was prepared by acid catalyzed dehydration of 3-hexen-2,5-diol.

Isomesityl oxide (III)'? was obtained by acid catalyzed enrichment of commercial mesityl oxide.

3-Methyl-4-penten-2-one (VI) and 5-hexen-3-one (1X)? were synthesized by reacting dicroty] and
diallyl zinc with ethyl cyanide.

3-Ethyl-3-buten-2-one (VII)! was prepared by the reaction of 2-ethyl-3-ketobutanoic acid with
formaldehyde and dimethylamine.

2-Ethylcyclobutanone (XVIID® was obtained by solvolysis of 3-hexyne-{-ol m-nitrobenzene-
sulfonate.

Cyclopropyl ethyl ketone (XXVI)"* and cyclopropyl acetone (XXVII)"* were isolated from the
reaction of cyclopropyl cyanide and cyclopropyl methyl cyanide with ethyl and methyl! magnesium
bromides respectively.

4-Methyl-4-penten-3-one (XD. A mixture of ethyl-2-methyl-3-oxovalerate (4-52 g, 28-6 mmole)
was stirred at room temperature with 1 N sodium hydroxide (30m!) for | hr. Dimethylamine
hydrochloride (2-5 g, 31-0 mmole) and 38% formaldehyde (3 ml) were added, the mixture heated to
85° for 1 hr., acidified with 5 N hydrochloric acid and extracted with ether. The ether extract on
evaporation and purification by g.l.c. afforded XI (5% yield). Mass spectrum, [M]* = mje 98; ir.
1680 cm~ (carbony]), n.m.r. in CDCI, (6 values) 1-05 (¢, J = 7 cps), 1-85 (d, J = 1 cps, 3H), 2°65
(q, J = 7 eps, 2H), 5-68 (g, J = | cps, 1H), 5:85 (s, 1H).

3-Ethylcyclobutyl methyl ketone. Methy] lithium (2-4 g, 0-11 m) in ether was added dropwise to a
cooled stirred solution of 3-ethylcyclobutane carboxylic acid’? (6-4 g, 0-05 m) in ether (10 ml). The
mixture was stirred for an additional 30 minutes and decomposed with a saturated ammonium chloride
solution. 3-Ethylcyclobutyl methyl ketone (4-5 g, 70% yield) was obtained as a colorless liquid
which was homogeneous by g.l.c. Mass spectrum [M]* = m/e 126; i-r. 1700 cm-? (carbonyl);
n.m.r. in CDCI, (6 values), 0-80 (¢, J = 7-0 cps, 3H), 1:40 c, 3H), 2:0 (s, superimposed on a complex
7H signal), 3-2 (c, 1H).

3-Ethylcyclobutyl acetate. This compound was obtained in 81% yield by Baeyer-Villiger oxi-
dation’® of 3-ethyleyclobutyl methyl ketone. The product was a colorless liquid and the following
physical properties were recorded. Mass spectrum, [M]* = m/e 142; ir. 1740 em! (carbonyl);
n.m.r. in CDCI, (6 values) 0-85 (1, J = 6°5 cps, 3H), 1:70 (c, SH), 2:0 (s, 3H), 2°58 (c, 2H), 4:85 (c, 1H).

3-Ethylcyclobutanol. 3-Ethylcyclobuty] acetate (5-0 g, 0-038 m) was hydrolyzed with 10% aqueous
potassium hydroxide for 5 hrs. at room temperature, 3-Ethylcyclobutanol was obtained as a colorless
liquid (3-35 g, 909% yield) which had the following properties: ir. 3350 cm! (hydroxyl), n.m.r. in
CDCI, (6 values) 0-85 (7, J = 6-5 cps, 3H), 1-40 (¢, 3H), 2-22(c, 5H).

3-Ethyleyclobutanone (XVII). A solution of sodium dichromate dihydrate (0°50 g, 1-68 mmoles)
in sulfuric acid (0-375 ml!) and water (2:0 ml) was added dropwise to a vigorously stirred solution of
3-ethylcyclobutanol! (0-5 g, 5 mmoles) in ether (8 ml). The mixture was vigorously stirred for 1-5 hrs.
After work up the ether layer afforded a pale liquid which contained 55% of 3-ethylcyclobutanone by
g.lic. The following physical properties were recorded for XVII: mass spectrum [M]~ = m/e 98; ir.
1780 cm-! (carbonyl); n.m.r. in CDCI, (6 values) 0-92 (t, J = 65 cps, 3H), 1-60 (¢, J = 6S cps, 3H),
2:50 (c, 2H), 3-0 (c, 2H).

1-Methyleyciopropyl methyl carbinol. A solution of 3-methyl-3-buten-2-ol (4-3 g, 0-05 m) in
anhydrous ether (10 ml) was added dropwise to the organozinc reagent prepared from zinc-copper
couple (4-0 g, 0-65 m) and diiodomethane (13-5 g, 0-05 m) in the usual manner. The adduct after
standing overnight was decomposed by 1N hydrochloric acid. Distillation of the product isolated
from the ether layer after work up and preparative g.1.c. afforded 1-methylcyclopropyl methyl] carbinol
(0-5, 10% yield) as a colorless liquid. The following physical data were obtained with this product:
ix. 3350 cm~? (hydroxyl), n.m.r. in CDCI, (6 values) 0-40 (c, 1-10 (c), 1-90 (c), 3-20 ¢, J = 6:5 cps).

1-Methyl-1-acetyleyclopropane (XXIV). 1-Methylcyclopropyl methyl carbinol (240 mg, 2:4
mmoles) was stirred with active manganese dioxide (5 g) in dichloromethane (50 ml) for 5 days. On
work up a pale liquid was obtained which on preparative g.l.c, furnished 1-methyl-1-acetyl cyclopro-
pane (XXIY) (70 mg, yield 30%). Mass spectrum [M]* = m/e 98; ix. 1693 cm-? (carbonyl), n.m.r.
in CDCI, (6 values) 0-70 (c, 2H), 1-20 (c, 2H), 1-35 (s, 3H), 2:04 (s, 3H).
Applications of artificial intelligence for chemical inference—V 501

Acknowledgements—Financial assistance from the Advanced Research Projects Agency (Contract
SD-183), the National Aeronautics and Space Administration (NGR-05-020-004) and the National
Institutes of Health (Grants GM 11309 and AM 04257) is gratefully acknowledged.

REFERENCES

1, Part IV. See A. Buchs, A. M. Duffield, G. Schroll, C. Djerassi, A. B. Delfino, B. G. Buchanan
G. L. Sutherland, E. A. Feigenbaum and J. Lederberg, J. Am. Chem. Soc., 92 (1970).

2, A. M. Duffield, A. V. Robertson, C. Djerassi, B. G. Buchanan, G. L. Sutherland, E. A. Feigen-
baum and J. Lederberg, J. Am. Chem. Soc. 91, 2977 (1969).

3. G.Schroll, A. M. Duffield, C. Djerassi, B. G. Buchanan, G. L. Sutherland, E. A. Feigenbaum and
J. Lederberg, J. Am. Chem. Soc. 91, 7440 (1969).

. J. Lederberg, Proc, Natl. Acad, Sci. U.S. 53, 134 (1965); J. Lederberg, ‘Topology of Molecules’ in
The Mathematical Sciences, M.1.T. Press, 1969, p. 37.

5. J. Lederberg, Am. Math. Monthly, 74, 522 (1965); J. Lederberg, DENDRAL-64, Part II,
‘Topology of Cyclic Graphs’, N.A.S.A. CR-68898, STAR No. N66-14074 (1965).

. J. Lederberg, G. L. Sutherland, B. G. Buchanan, E. A. Feigenbaum, A. V. Robertson, A. M.
Duffield and C. Djerassi, J. Am. Chem. Soc. 91, 2973 (1969).

7. J. Kossanyi, Bull. Soc. Chim. France 704 (1965).

8. D. S. Weiss, R. B. Gargosian and N. J. Turro, Org. Mass Spectrom, 3, 145 (1970).

9. J. L. Pierre, R. Barlet and P. Arnaud, Spectrochim, Acta 23, 2297 (1967).

10. J. L. Pierre and P. Arnaud, Bull, Soc, Chim. France 1690 (1966).

11. H. Morrison, Tetrahedron Letters 3653 (1964).

12. R. Mecke and K. Noack, Chem. Ber. 93, 210 (1960).

13. M. Gaudemar, Bull. Soc. Chim. France 974 (1962).

14. S. Szantay and J. Rohdly, Chem. Ber. 96, 1788 (1963).

15. M. Hanack and I. Herterich, Tetrahedron Letters 3847 (1966).

16. M. Hanack and H. M. Ensselin, Annalen 697, 100 (1966).

17. K. C. Brannoek, A. Bell, R. D. Burpitt and C. A. Kelly, J. Org. Chem. 29, 801 (1964).

18. W. D. Emmons and G. B. Lucas, J. Am. Chem. Soc. 77, 2287 (1955).

>

a