facebook | Read more | Анатолий Папанов

Symmetric vertex models on planar random graphs

Symmetric vertex models on planar random graphs

23 September 1999 Physics Letters B 463 Ž1999. 9–18 Symmetric vertex models on planar random graphs D.A. Johnston Dept. of Mathematics, Heriot-Watt ...

108KB Sizes 1 Downloads 18 Views

23 September 1999

Physics Letters B 463 Ž1999. 9–18

Symmetric vertex models on planar random graphs D.A. Johnston Dept. of Mathematics, Heriot-Watt UniÕersity, Riccarton, Edinburgh, EH14 4AS, Scotland, UK Received 27 July 1999; accepted 11 August 1999 Editor: P.V. Landshoff

Abstract We solve a 4-Žbond.-vertex model on an ensemble of 3-regular ŽF 3 . planar random graphs, which has the effect of coupling the vertex model to 2D quantum gravity. The method of solution, by mapping onto an Ising model in field, is inspired by the solution by Wu et.al. of the regular lattice equivalent – a symmetric 8-vertex model on the honeycomb lattice, and also applies to higher valency bond vertex models on random graphs when the vertex weights depend only on bond numbers and not cyclic ordering Žthe so-called symmetric vertex models.. The relations between the vertex weights and Ising model parameters in the 4-vertex model on F 3 graphs turn out to be identical to those of the honeycomb lattice model, as is the form of the equation of the Ising critical locus for the vertex weights. A symmetry of the partition function under transformations of the vertex weights, which is fundamental to the solution in both cases, can be understood in the random graph case as a change of integration variable in the matrix integral used to define the model. Finally, we note that vertex models, such as that discussed in this paper, may have a role to play in the discretisation of Lorentzian metric quantum gravity in two dimensions. q 1999 Elsevier Science B.V. All rights reserved.

1. Introduction Random matrix models have proved remarkably successful in investigating the critical behaviour of various sorts of spin models ŽIsing w1,2x, Potts w3x, O Ž N . w4x . . . . living on ensembles of planar random graphs. This, in effect, couples the models to 2 D quantum gravity which originally motivated their study in the context of string theory and random surfaces w5x. From the purely statistical mechanical point of view one is looking at the critical behaviour of an Žannealed. ensemble of random graphs decorated by the spins. The connection with gravity comes from the fact that the graphs can adapt their connectivity in response to the spin configuration, which in turn is influenced by the connectivity of the

graph on which it lives. This captures the essential feature of the back reaction of the matter on the geometry that characterises gravity in a discrete form. The graphs of interest appear as planar Feynman diagrams in the perturbation expansion in the vertex coupling of a Hermitian matrix model free energy of the general form w6x

Fs

1 M2

H Łi DF exp Ž yS .

log

i

Ž 1.

in the limit M ™ `, where the F i are sufficient M = M Hermitian matrix variables to describe the matter decoration and the log ensures that only connected graphs contribute. For the Ising model on

0370-2693r99r$ - see front matter q 1999 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 0 - 2 6 9 3 Ž 9 9 . 0 0 9 4 8 - X

D.A. Johnstonr Physics Letters B 463 (1999) 9–18

10

trivalent planar graphs, for instance, we need two matrices X,Y and the action takes the form S s Tr

½

1 2

l y 3

Ž X 2 q Y 2 . y gXY

5

w e h X 3 q eyh Y 3 x ,

Ž 2.

where g s expŽy2 b . and h is the external field. We can see that the X 3 terms can be thought of as spin-up or q vertices and the Y 3 terms spin-down or y, while the inverse of the quadratic terms gives the appropriate edge weights for qq, qy and yy edges. To date there has been less consideration of arrow vertex models on ensembles of planar random graphs and none at all of bond vertex models. In the arrow case one decorates the edges of the graph with arrows and assigns weights to the vertices depending on the local arrow configurations around each vertex, in the bond case one simply has empty or occupied edges Žbonds. and vertex weights depending on the local bond configuration. Some arrow vertex models have been discussed using elementary means in w7,8x, mostly on 4-valent random graphs, revealing XY and Ising critical behaviour for suitable coupling constant loci. More recently some elegant and sophisticated work using character expansions w9x has provided a full solution of an arrow vertex model on 4-valent planar graphs confirming the XY critical point found in w7x. In this paper we give the first solutions of bond vertex models on planar random graphs. Our main example is a 4-vertex model on F 3 random graphs, but the general method works for any bond vertex models whose weights depend only on the number of bonds at a vertex rather than their particular cyclic ordering. On regular lattices such models are termed symmetric vertex models and we retain the nomenclature here. Since our 4-vertex model on F 3 random graphs is the random graph equivalent of the symmetric 8-vertex model on the honeycomb lattice discussed by Wu et al. w10–12x we take our cue from w10x where the honeycomb lattice model was first solved by making use of the so-called generalised weak graph transformation w13x between vertex weights. This brought the partition function to a form that was recognisable as that of an Ising model in

field an the critical behaviour of the model was then deduced from the corresponding Ising results. In what follows we first define our 4-vertex model, then solve it by mapping onto the Ising model in field on F 3 planar random graphs, which itself was solved in w1x. We move on to discuss higher valency symmetric models and show that the same general method of solution still works. We then compare these results with those of w10–12x for the 8-vertex model on the honeycomb lattice and note that the generalized weak graph symmetry of the honeycomb lattice model finds a counterpart in an orthogonal variable transformation in the matrix integrals used to define the random graph models. Finally, we close with some speculations on the relations of vertex models to Lorentzian metric quantum gravity, where they can be thought of as providing a discretisation of causal structure.

2. The 4-vertex model on 3-valent random graphs The canonical Žfixed number of vertex. partition function for the 4-vertex model on F 3 planar random graphs is given by Z4 Õ s Z4 Õ Ž a,b,c,d . s Ý

Ý aN b N 1

2

c N 3 d N4 ,

Ž 3.

FN G

where there are two summations: ÝF N over F 3 random graphs with N vertices and ÝG over all possible bond configurations on each F 3 graph built using the vertices of Fig. 1. Each graph has N1 vertices of type Ž a., N2 of type Ž b ., N3 of type Ž c ., N4 of type Ž d . and N s N1 q N2 q N3 q N4 . We can generate the graphs we require, suitably decorated with the allowed bond configurations, by inserting the action S s Tr

½

yl

1 2

Ž X 2qY 2 . a 3

X 3q

d 3

Y 3 q cXY 2 q bX 2 Y

5

.

Ž 4.

into Eq. Ž1., where X,Y are M = M Hermitian matrices and we have introduced an overall vertex coupling l for convenience.. The limit M ™ ` picks out the planar diagrams. A direct attempt at solving a matrix model with the action Eq. Ž4. might be possi-

D.A. Johnstonr Physics Letters B 463 (1999) 9–18

11

weights from Eq. Ž6. are also given in Fig. 1. It should be remarked that the transformation of Eq. Ž5. is the duality transformation for the Ising model on random F 3 graphs w14x, and that the bond graphs generated by the transformed action in Eq. Ž6. are precisely the high-temperature expansion graphs of the original Ising model on F 3 graphs. We can see that the vertex weights coming from the Ising model in Eq. Ž6., which we denote by tilde’d quantities, b˜ s sinh Ž h . Ž g ) .

a˜ s cosh Ž h . ,

d˜s sinh Ž h . Ž g .

c˜ s cosh Ž h . g ,

ble using the character expansion techniques of w9x, but it turns out to be easier to follow in the footsteps of w10,11x and establish a correspondence with an Ising model in order to determine the critical behaviour of the vertex model. We start by by carrying out the orthogonal transformations X ™ Ž X q Y . r'2 ,

Y ™ Ž X y Y . r'2

Ž 5.

on the matrices in the Ising action of Eq. Ž2., followed by the rescalings X ™ XrŽ1 y g .1r2 , Y ™ YrŽ1 q g .1r2 , l ™ '2 lŽ1 y g . 3r2 . This change of variable in the matrix integral has a trivial Jacobian and gives the new action S s Tr

y

½

1 2

2

Ž X qY .

3 y

X 3 q 3 g ) XY 2

lsinh Ž h . Ž g ) . 3

3r2

Y 3q

3 g)

X 2Y

5

Ž 7.

˜ ˜ which will not be the are fixed in the ratio ad ˜ ˜s bc, case for the generic weights of Eq. Ž4.. However, if we consider the effect of the further orthogonal rotation 1 cos Ž u . X ™ Y ysin Ž u .

ž /

ž

sin Ž u . cos Ž u .



X Y

/

Ž 8.

on the vertex weights of Eq. Ž4., we find 1 a˜ s a q 3 yb q 3 y 2 c q y 3d , 3r2 Ž1 q y 2 . b˜ s

1 2 3r2

yya q Ž 1 y 2 y 2 . b

Ž1 q y . yŽ y 3 y 2 y . c q y 2d c˜ s

d˜s

1 3r2

y 2a q Ž y 3 y 2 y . b

Ž1 q y 2 . q Ž 1 y 2 y 2 . c q yd 1 2 3r2

,

,

yy 3a q 3 y 2 b y 3 yc q d ,

Ž 9.

Ž1 q y .

where we have extracted cosŽ u ., denoted tanŽ u . as y and deliberately used tilde’d quantities again for the transformed weights. We can now demand that ˜ ˜ just as in these transformed weights satisfy ad ˜ ˜s bc, Eq. Ž6. which has the effect of making the partition function of the vertex model equivalent to that of the ˜ ˜ gives the folIsing model in field. Setting ad ˜ ˜s bc

2

lcosh Ž h .

,

) 3r2

)

Fig. 1. The possible bond vertices which appear in the model, using the same notation as in w10x. The Ising weights on the random lattice, which can be read off from Eq. Ž6., are: as ˜ cosh Ž h ., b˜ s sinhŽ h .Ž g ) .1r 2 , c˜ s coshŽ h . g ) and d˜ s sinhŽ h.Ž g ) . 3r 2 .

1r2

,

Ž 6.

where g ) s Ž1 y g .rŽ1 q g ., which is clearly of the same general form as Eq. Ž4.. The bond vertex

1 This preserves the propagator X 2 qY 2 in the action of Eq. Ž6. and again has a trivial Jacobian, so it does not effect the evaluation of the matrix integral.

D.A. Johnstonr Physics Letters B 463 (1999) 9–18

12

lowing equation for the rotation parameter y from equating the right hand sides of Eq. Ž9.

critical temperature is given by bc s 12 log Ž 103 28 . s 0.7733185 w2x. The first of Eqs. Ž12. then gives

By 2 q 2 Ž C y A . y y B s 0 ,

Ž 10 .

103

where A s bd y c 2 , B s ad y bc, C s ac y b 2 . Since y` F y F ` is at our disposal we can solve this for general a,b,c,d to obtain transformed weights ˜˜ satisfying ad ˜ ˜s bc. To deduce the critical behaviour of the vertex model by using this correspondence with the Ising model in field we need the relation between the vertex weights and the Ising parameters b and h. The Ising and tilde’d vertex model parameters were related by

28



tanh Ž b . s



,

tanh Ž h . s



'ac ˜˜

.

Ž 11 .

as can be seen directly from Eq. Ž6.. We can then substitute for the tilde’d weights in terms of the original weights and y using Eq. Ž9.. to give the relation between the original weights a,b,c,d and the Ising parameters b ,h exp Ž 2 b . s

tanh Ž h . s

By q C y A

W T

AqC

ž

By q 2C By y 2 A

, 1r2

/

,

Ž 12 .

where y is again the one of the solutions of Eq. Ž10., A, B,C are as above, T s Ž b q d . y q a q c and W s Ž b q d . y Ž a q c . y. On F 3 random graphs the Ising model displays a third order magnetisation transition in zero field. The second of Eqs. Ž12. shows that zero Ising field implies that W s 0, which in terms of the vertex weights Žwhen y is substituted for. gives the locus 3

3

3

(Ž ac y b y bd q c . q Ž ad y bc . s

Ž 14 . as the position of the third order critical point on the locus Eq. Ž13.. We still have, of course, the echo of the field driven Ising transition appearing for b ) bc along the locus Eq. Ž13.. 3. Higher valency vertex models In this section we show that on 4-valent ŽF 4 . random graphs the Ising equivalency can still be easily established by rotating the variables in a matrix model action. One extra ingredient in the 4-valent case compared with the 3-valent case is that the cyclic order of the bonds allows one to distinguish between two different sorts of vertices with two occupied and two empty edges, one with occupied bonds at right angles and one with occupied bonds ‘‘straight through’’. The Ising equivalence only holds for the symmetric case in which equal weights are assigned to these bond configurations. On F 4 random graphs the Ising model action is 1 S s Tr Ž X 2 q Y 2 . y gXY 2 l q w e h X 4 q eyh Y 4 x , Ž 15 . 4 with only a higher order potential distinguishing the action from that for F 3 graphs in Eq. Ž2.. The change of variables X ™ Ž X q Y .r '2 , Y ™ Ž X y Y .r '2 and rescalings X ™ XrŽ1 y g .1r2 , Y ™ YrŽ1 q g .1r2 , l ™ 2 lŽ1 y g . 2 gives the action 1 S s Tr Ž X 2qY 2 . 2

½

5

½

Ž 13 . From the vertex model perspective it is most natural to disallow loops of length one and two in the random graphs in order to make sure that there are no double edges or tadpoles, in which case the Ising

lcosh Ž h .

2

X4qŽ g). Y 4 4 lcosh Ž h . g ) w 2 X 2 Y 2 q XYXY x q 2 1 qlsinh Ž h . g ) g ) XY 3 q X 3Y g) q

q 3 Ž ab q bc q cd . Ž c 2 y bd y b 2 q ac . s 0 .

2

ac y b 2 q bd y c 2

3

aŽ b q d . y d Ž a q c .

2 2

2

'

'

5

,

Ž 16 .

D.A. Johnstonr Physics Letters B 463 (1999) 9–18

e˜ s

1

Ž1 q y 2 .

2

13

y 4 a y 4 y 3 b q 6 y 2 c y 4 yd q e .

Ž 18 .

Fig. 2. The 5 different vertex weights in the symmetric 16 vertex model on the square lattice. The ‘‘straight-through’’ two bond configuration which is not shown also receives weight c.

where, as on F 3 graphs, g ) s Ž1 y g .rŽ1 q g .. Note the presence of the w 2 X 2 Y 2 q XYXY x term fixing the weights of right angled and straight through two-bond vertices to be the same. It should be remarked once again that the change of variables in Eq. Ž5. is a duality transformation for the model, and that Eq. Ž16. generates the high temperature expansion graphs of the original Ising model in Eq. Ž15.. The effect of the orthogonal rotation of Eq. Ž8. on the vertices arranged in some suitable lexicographic order Že.g. X 4 , X 3 Y, X 2 Y 2 , XY 3 , Y 4 as labelled on Fig. 2. gives 1 a˜ s a q 4 yb q 6 y 2 c q 4 y 3d q y 4 e , 2 2 Ž1 q y . b˜ s

1 2 2

yya q Ž 1 y 3 y 2 . b

Ž1 q y . y Ž3 y 3 y 3 y . c q Ž3 y 2 y y 4 . d q y 3e c˜ s

1

Ž1 q y 2 .

2

,

y 2 a q Ž2 y 3 y 2 y . b

q Ž y 4 y 4 y 2 q 1. c q Ž2 y y 2 y 3 . d q y 2 e , d˜s

1

yy 3a q Ž 3 y 2 y y 4 . b 2 Ž1 q y 2 . y Ž 3 y y 3 y 3 . c q Ž 1 y 3 y 2 . d q ye

,

Ž 17 .

The strategy for solving the F 4 model is identical to the F 3 case. One uses the above transformations of the vertex weights to bring the model to the Ising action defined above in Eq. Ž16.. The tilde’d weights in this Ž a˜ for X 4 etc.. can be characterised in various ˜˜ c˜ 2 s ae. ways, for example: ad ˜ ˜s dc, ˜˜ Having achieved this the critical behaviour can again be read off from that of the Ising model in field, this time on F 4 graphs. Indeed, it is clear that generic symmetric vertex models on random graphs can be solved by using this method. In the q-valent case one carries out the same sequence of two orthogonal rotations on the action for the Ising model on F q planar random graphs S s Tr

½

1 2

l q q

Ž X 2 q Y 2 . y gXY

5

w e h X q q eyh Y q x .

Ž 19 .

to obtain the action for the symmetric vertex model. One can thus deduce that the critical behaviour of the symmetric vertex model on q-valent random graphs will be Ising-like.

4. Comparison with the symmetric 8-vertex model on the honeycomb lattice The closest regular lattice equivalent of our F 3 planar random graphs is the honeycomb lattice where each vertex still has coordination number three but, in contrast to the random graphs, all loops are of length six. The eight possible bond vertex configurations all look similar to those for the random lattice in Fig. 1 but we now have orientational order as well as cyclic order around each vertex so b and c represent three possible orientations each. However if we impose equality between the different orientations, we arrive at the symmetric vertex model of w10–12x which, like the F 3 random graph vertex model, has 4 distinct vertex weights. The partition

D.A. Johnstonr Physics Letters B 463 (1999) 9–18

14

function is given in a similar fashion to Eq. Ž3. by Z8 Õ s Z8 Õ Ž a,b,c,d . s Ý a N1 b N 2 c N3 d N4 ,

Ž 20 .

G

where there is now no sum over random graphs since all the bond configurations are living on the honeycomb lattice. Several symmetry properties of the partition function are immediately apparent. From the negation of vertices occurring in pairs we have Z8 Õ Ž a,b,c,d . s Z8 Õ Žya,y b,y c,y d . s Z8 Õ Žya,b,y c,d . s Z8 Õ Ž a, yb,c,y d .. Similarly, exchanging dark and light bonds gives Z8 Õ Ž a,b,c,d . s Z8 Õ Ž d,c,b,a.. Perhaps rather less obvious is the generalised weak graph symmetry w13x Z8 Õ Ž a,b,c,d . s Z8 Õ Ž a,b,c,d ˜ ˜ ˜ ˜., where the transformations between the original and tilde’d variables are precisely those generated by the orthogonal rotations in the random graph model in Eq. Ž9.. In w11x the generalised weak graph symmetry on the honeycomb lattice was thought of as being generated by rotations with V Ž y. s

1 2 1r2

Ž1 q y .

ž

1 y

y y1

/

Ž 21 .

on the vertices in ‘‘bond space’’, where occupied and empty bonds on a given edge were to be thought of as components of a vector. An alternative choice of transformation UŽ y. s

1 2 1r2

Ž1 q y .

ž

1 yy

y 1

/

Ž 22 .

where Z Ising was the standard honeycomb lattice Ising partition function. The equivalence followed from observing that this choice of weights meant that the vertices generated precisely the right weights for the diagrams of the high temperature expansion of the Ising model on the honeycomb lattice. The critical behaviour of the symmetric 8-vertex model on the honeycomb lattice was thus deduced by following exactly the same path we have taken for the 4-vertex model on F 3 random graphs – exploiting a symmetry of the partition function to transform it to that of an Ising model in field. The parallels run even closer since Eqs. Ž9. – Ž13. still hold identically for the honeycomb lattice vertex model and the only difference in Eq. Ž14. is in the numerical value of the Ising critical temperature on the left hand side. This might grandiloquently be phrased as a non-renormalization theorem for the vertex weights since the ratios of weights in Eq. Ž11. are unaffected by the coupling to gravity that is represented by the sum over random graphs. The close correspondence between the F 3 and honeycomb results becomes a little less surprising when one considers a second determination of the critical behaviour of the honeycomb lattice model by Wu in w12x where it was shown that a direct mapping between the Ising and vertex models was possible when a decoration-iteration transformation was employed. In Fig. 3 we show a single vertex, which hosts a spin subject to an external field H. In addition the edges have spins which are subject to a

˜ was related to V Ž y . by the negations b˜ ™ yb, d˜™ yd˜ and was hence not independent. Remarkably, the second transformation is just that generated by Eq. Ž8. in the random graph model. The generalised weak graph symmetry played an important role in the original solution of the honeycomb lattice model w11x because a suitable choice of y allowed generic vertex weights a,b,c,d to be ˜ ˜ just as for transformed to a,b,c,d ˜ ˜ ˜ ˜ satisfying ad ˜ ˜s bc the random lattice. For vertex weights satisfying the latter condition Z8 Õ Ž a,b,c,d Ž h . r2 . ˜ ˜ ˜ ˜ . s Ž acosh ˜ = Ž cosh Ž b . .

N

y3 Nr2

Z Ising Ž b ,h . ,

Ž 23 .

Fig. 3. The decorating spins on the edges are subject to an X external field 2 H , the central spin to a field H. The edge-vertex spin interactions have weight R.

D.A. Johnstonr Physics Letters B 463 (1999) 9–18

field 2 H X and the two sorts of spins interact via a spin-spin coupling R. The presence of a dark bond can be denoted by an edge spin s s 1 and its absence by s s y1. The vertex model weights can then be represented correctly in terms of appropriate H X , H and R cosh Ž 2 R . s

cosh Ž 2 H . s

B 2 Ž AC . 2 bc

'AC

ž

1r2

,

X

exp Ž 4 H . s

B2

B y

4 AC

4 bc

/

y1 ,

C A

duced by using the appropriate generalized weak graph transformations to map the models onto Ising models in field in all cases. The generalised weak graph transformation written down by Wu for general valency q Wi j s

,

15

j

1 2 qr2

Ž1 q y .

Ý ks0

i k

qyi jyk

ž / ž /Ž

k

y1 . y iqjy2 k ,

Ž 25 . Ž 24 .

when the central vertex spins are traced over. One can go in the other direction and decimate the edge Ising spins by replacing the two R interactions and edge field 2 H X with a single interaction b and fields at the end of each edge Ži.e. on the vertices.. This gives one a standard honeycomb lattice Ising model in field. The relation between the vertex model weights and the Ising parameters b ,h determined in this manner is identical to that Eq. Ž12.. This decoration-iteration approach strongly supports the observation here that the equivalence between the symmetric 8-vertex model on the honeycomb lattice and the Ising model in field continues to hold for the 4-vertex model on F 3 random graphs. The decoration iteration transformation is a local transformation which relies only on the valency of each vertex and in both cases all vertices have valency three. It also supports the result that the relation between the Ising and vertex parameters is identical on F 3 random graphs and the honeycomb lattice, since the decoration-iteration transformation doesn’t ‘‘see’’ the randomness, so as far as it is concerned the random and honeycomb lattices are identical 2 . The critical behaviour of symmetric vertex models on higher valency regular lattices may be de-

2 One possible fly in the ointment is that loops of length one and two, i.e. double edges Ž‘‘bubbles’’. and self-loops Ž‘‘tadpoles’’., are in principle present in the random graphs. The star-triangle relation would degenerate at vertices adjacent to such edges. The simplest solution, which we have taken, is to restrict the ensemble of random graphs to exclude such configurations, which does not affect the critical behaviour of the Ising model though non-universal quantities such as the critical temperature may be changed.

where the vertices are labelled in some suitable order as in the F 3 , F 4 cases, is precisely that obtained by picking out the appropriate powers in an expansion of terms such as ŽcosŽ u . X q sinŽ u .Y . qy k ŽysinŽ u . X q cosŽ u .Y . k .

5. Speculations on connections with Lorentzian gravity

F 3 graphs are duals to triangulations which are the natural discretisation of Euclidean signature Žqq . 2D spacetimes. An analytical continuation to physical Lorentzian Žyq . signatures is by no means on such firm ground as the analogous procedure in standard quantum field theory on flat backgrounds. It is therefore of some interest to attempt to formulate discretised theories which might serve as toy models for Lorentzian gravity. An essential extra ingredient compared with the Euclidean case for any Lorentzian theory is a causal structure. One then faces the option of whether causal structures should be summed over or imposed. A relatively rigid choice was explored in w15x where triangulations with spacelike slices joined by fluctuating timelike edges were used to triangulate Lorentzian space. It was found that only when branchings were allowed was the critical behaviour of Euclidean gravity recovered upon analytical continuation back. Another possibility is to postulate that global, or at least long-range, causal structure might emerge dynamically in much the same way that the fractal structure of the 2D manifolds does in the Euclidean case. Inspired by w16,17x we introduce local causal structure by using spacelike edges to triangulate our Lorentzian spacetime. The normals to all the triangles are then time-like, which can be either past or

D.A. Johnstonr Physics Letters B 463 (1999) 9–18

16

future directed. It is then natural to return to the dual spin network picture, which gives an embedded directed F 3 graph if we denote the causal relations between points in the interior of adjacent triangles by arrows pointing from the past to the future of each vertex. This should be contrasted with Euclidean 2D triangulations where the dual picture is of undirected F 3 graphs. It is thus natural to use arrow vertex models in this context, rather than the bond vertex models considered so far. However the translation between arrows and bonds turns out to be trivial, as we see below. The various sorts of 3-valent ŽF 3 . arrow vertices are most conveniently labelled by their Žindegree, outdegree.. The different possibilities are Ž3,0., Ž0,3., Ž2,1. and Ž1,2. as shown in Fig. 4. A natural choice is to take conjugate weights for the Ž3,0. and Ž0,3. vertices and similarly for the Ž2,1. and Ž1,2. vertices as this preserves the symmetry under arrow reversal. The partition function for this 4-arrow-vertex model is thus of the form Zs Ý

Ý aN aN 1

2

b N 3 b N4 ,

Ž 26 .

FN G

where the first sum is again over different planar F 3 graphs, the second over arrow assignments and we

have N1 Ž3,0. vertices, N2 Ž0,3. vertices etc. Note that only the Ž2,1. and Ž1,2. vertices can be considered as regular spacetime points since the Ž3,0. and Ž0,3. vertices act as microscopic black and white holes respectively. We can obtain this partition function from the N vertex term in the expansion of the free energy of the complex matrix model with the action Ss

1 2

½

Tr F †F y

b yi 3

a

3

ŽF † . qF 3

3

3

2

ŽF 3 y ŽF † . yg ŽF † . FqF 2 ŽF † . 2

yi d Ž F † . F y F 2 Ž F † .

5

,

Ž 27 .

where F is now an M = M complex matrix and a s a q i b , b s g q i d . We can transform simply to a bond vertex formulation of the model by splitting F into Hermitian components X q iY. This allows us to interpret Y edges as those containing bonds and X edges as empty Žor Õice-Õersa.. The resulting action in terms of X,Y S s Tr y

½

1 2

Ž X 2qY 2 . y

Ž b q 3d . 3

Ž a q 3g . 3

X3

Y 3 y Ž g y a . XY 2

5

yŽ d y b . X 2 Y .

Ž 28 .

is now clearly that of our 4-bond-vertex model on F 3 random graphs. This displays Ising criticality, as we have seen, so a suitable tuning of the vertex couplings in Eq. Ž28. Žand hence in the original arrow vertex model of Eq. Ž27.. can reach this point 3. Since this is a continuous transition one might hope that the diverging correlations of the Ising model at the critical point might be translated back to the appearance of long range casual structure ex nihilo in the arrow vertex model formulation. However, one can see heuristically that closed loops

Fig. 4. The possible arrow vertices which appear in the model. Their weights are equivalent to linear combinations of the bond vertex weights in Fig. 1. On the random lattices the corresponding terms in the action are:Ža. ŽF † . 3 , Žb. F 3 , Žc. ŽF † . 2F , Žd. F 2F †

3 Interestingly, the presence of the black hole and white hole vertices appears to be necessary to do this, there is insufficient freedom with only the regular vertices.

D.A. Johnstonr Physics Letters B 463 (1999) 9–18

of arrows are present at all scales at the transition point. Cluster algorithms exist for vertex models w18x which act by identifying such closed loops and flipping them. These are effective at combatting critical slowing down so they are identifying and flipping loops at all scales near a continuous transition point. The loops are innocuous from the statistical mechanical point of view, but they represent closed timelike loops when the vertex arrows are interpreted as giving the causal structure. A profusion of closed timelike loops is obviously an undesirable property and we thus conclude that the particular vertex model discussed here is probably not a suitable candidate for modeling discretised Lorentzian gravity as it stands. It would be interesting to determine what modifications might make it so.

6. Discussion We have solved a bond 4-vertex model on an ensemble of planar F 3 graphs by taking our lead from the honeycomb lattice solution of the symmetric 8-vertex model and exploiting its equivalence to an Ising model in field. An important ingredient of the solution was an orthogonal rotation in the matrices use to define the random graph model in order bring a generic weight configuration on to the Ising locus. This turned out to be functionally identical to the generalised weak graph transformation used in the original honeycomb lattice solution w10x. The method of solution and the equivalence of the matrix rotation to the generalised weak graph transformation on an equivalent regular lattice were also shown to work for higher valency symmetric vertex models on random graphs. We have seen that the relations between the vertex model parameters and the Ising parameters were identical for the honeycomb and random graph models. This is not so surprising at it might first seem when viewed in the light of the decoration-iteration solution on the honeycomb lattice, since this depends only on the valency of the vertices, which is identical in the random case Žmodulo caveats about tadpoles and bubbles discussed previously.. This result could be couched as a non-renormalization theorem for the appropriate ratios of vertex weights, since putting the

17

models on ensembles of random graphs is equivalent to coupling them to 2D quantum gravity. The methods discussed in this paper are restricted to symmetric vertex models since the orthogonal rotation of Eq. Ž5. necessarily gives only symmetric weights when applied to Ising actions such as those in Eqs. Ž2., Ž15. and Ž19.. From the matrix model point of view the vertex model solution shows that an apparently hopeless potential containing X 3, Y 3 , X 2 Y and XY 2 terms can still give rise to a soluble model, once the matrices are appropriately transformed. Finally, we have speculated on the relation between vertex models and Lorentzian signature gravity, pointing out a potential problem with closed timelike loops when employing the class of models considered here.

Acknowledgements This work was partially supported by a Leverhulme Trust Research Fellowship and a Royal Society of EdinburghrSOEID Support Research Fellowship.

References w1x V. Kazakov, Phys. Lett. A 119 Ž1986. 140; D. Boulatov, V. Kazakov, Phys. Lett. B 186 Ž1987. 139. w2x Z. Burda, J. Jurkiewicz, Acta Physica Polonica B 20 Ž1989. 94. w3x V. Kazakov, Mod. Phys. Lett. A 4 Ž1989. 1691; in Seillac Symposium Ž1987. 93. w4x B. Eynard, C. Kristjansen, Nucl. Phys. B 516 Ž1998. 529; B 455 Ž1995. 577; B. Eynard, J. Zinn-Justin, Nucl. Phys. B 386 Ž1992. 558; B. Duplantier, I. Kostov, Phys. Rev. Lett. 61 Ž1988. 1433; I. Kostov, Mod. Phys. Lett. A 4 Ž1989. 217; M. Gaudin, I. Kostov, Phys. Lett. B 220 Ž1989. 200. w5x V. Knizhnik, A. Polyakov, A. Zamolodchikov, Mod. Phys. Lett. A 3 Ž1988. 819; F. David, Mod. Phys. Lett. A 3 Ž1988. 1651; J. Distler, H. Kawai, Nucl. Phys. B 321 Ž1989. 509. w6x E. Brezin, C. Itzykson, G. Parisi, J. Zuber, Commun. Math. Phys. 59 Ž1978. 35; M. Mehta, Commun. Math. Phys. 79 Ž1981. 327. w7x P. Ginsparg, Matrix Models and 2D Gravity, hepthr9112013, also in Phys. Rep. 254 Ž1995.. w8x D. Johnston, P. Plechac, Phys. Lett. A 248 Ž1998. 37. w9x Two-Matrix model with ABAB interaction, V. Kazakov, P. Zinn-Justin, hep-thr9808043. w10x F. Wu, J. Math. Phys. 15 Ž1974. 687; F. Wu, X. Wu, H.

18

w11x w12x w13x w14x

w15x

D.A. Johnstonr Physics Letters B 463 (1999) 9–18 Blote, ¨ Phys. Rev. Lett. 62 Ž1989. 2773; X. Wu, F. Wu, J. Phys. A 22 Ž1989. L55. X. Wu, F. Wu, J. Stat. Phys. 50 Ž1989. 41. F. Wu, J. Phys. A 23 Ž1990. 375. F. Wegner, Physica 68 Ž1973. 570; J. Nagle, J. Math. Phys. 8 Ž1968. 1007. S. Carroll, M. Ortiz, W. Taylor, Nucl. Phys. B 468 Ž1996. 383; Nucl. Phys. B 468 Ž1996. 420; Phys. Rev. Lett. 77 Ž1996. 3947; D. Johnston, Phys. Lett. B 314 Ž1993. 6. J. Ambjørn, R. Loll, Nucl. Phys. B 536 Ž1998. 407; A new perspective on matter coupling in 2d quantum gravity, J. Ambjørn, K. Anagnostopoulos, R. Loll, hep-thr9904012; J.

Ambjørn, J. Nielsen, R. Loll, Chaos Solitons Fractals 10 Ž1999. 177. w16x F. Markopoulou, L. Smolin, Phys. Rev. D 58 Ž1998. 084032. w17x Dual formulation of spin network evolution, F. Markopoulou wgr-qcr9704013x w18x H. Evertz, G. Lana, M. Marcu, Phys. Rev. Lett. 70 Ž1993. 875; H. Evertz, M. Marcu, Nucl. Phys. Proc. Suppl. 30 Ž1993. 277; H. Evertz, in: D.J. Scalapino ŽEd.., Numerical Methods for Lattice Quantum Many-Body Problems, Addison Wesley Longman, Frontiers in Physics; G. Barkema, M. Newman, Phys. Rev. E 57 Ž1998. 1155.