Solution to a problem arising from Mayer’s theory of cluster integrals Olivier Bernardi, C.R.M. Barcelona October 2006, 57th Seminaire Lotharingien de Combinatoire CRM, Barcelona Olivier Bernardi – p.1/37 Content of the talk Mayer’s theory of cluster integrals Pressure = Generating function of weighted connected graphs. CRM, Barcelona OOlivier Bernardi – p.2/37 Content of the talk Mayer’s theory of cluster integrals Pressure = Generating function of weighted connected graphs. Hard-core continuum gas Pressure = Generating function of Cayley trees. CRM, Barcelona OOlivier Bernardi – p.2/37 Content of the talk Mayer’s theory of cluster integrals Pressure = Generating function of weighted connected graphs. Hard-core continuum gas Pressure = Generating function of Cayley trees. Why ? [Labelle, Leroux, Ducharme : SLC 54] CRM, Barcelona OOlivier Bernardi – p.2/37 Content of the talk Mayer’s theory of cluster integrals Pressure = Generating function of weighted connected graphs. Hard-core continuum gas Pressure = Generating function of Cayley trees. Why ? [Labelle, Leroux, Ducharme : SLC 54] Combinatorial explaination CRM, Barcelona Olivier Bernardi – p.2/37 Mayer’s theory of cluster integrals CRM, Barcelona Olivier Bernardi – p.3/37 Statistical physics Gas of n particules in a box Ω. x2 Ω x1 CRM, Barcelona x3 OOlivier Bernardi – p.4/37 Statistical physics Gas of n particules in a box Ω. x2 Ω x1 x3 The energy of a configuration x1 , . . . , xn is X X (x1 , . . . , xn ) = µ(xi ) + φ(xi , xj ). i CRM, Barcelona i<j OOlivier Bernardi – p.4/37 Statistical physics Gas of n particules in a box Ω. x2 Ω x1 x3 The energy of a configuration x1 , . . . , xn is X X (x1 , . . . , xn ) = µ(xi ) + φ(xi , xj ). i i<j No external field : µ(xi ) = µ. CRM, Barcelona Olivier Bernardi – p.4/37 Statistical physics x2 Ω x3 x1 Energy : (x1 , . . . , xn ) = nµ + CRM, Barcelona P i<j φ(xi , xj ). OOlivier Bernardi – p.5/37 Statistical physics x2 Ω x3 x1 Energy : (x1 , . . . , xn ) = nµ + P i<j φ(xi , xj ). The partition function is ZZ 1 (x1 , . . . , xn ) Z(Ω, T, n) = exp − dx1 ..dxn . n! Ωn kT CRM, Barcelona OOlivier Bernardi – p.5/37 Statistical physics x2 Ω x3 x1 Energy : (x1 , . . . , xn ) = nµ + P i<j φ(xi , xj ). The partition function is ZZ 1 (x1 , . . . , xn ) Z(Ω, T, n) = exp − dx1 ..dxn n! Ωn kT ZZ Y 1 φ(xi , xj ) = n exp − dx1 ..dxn . λ n! Ωn i<j kT CRM, Barcelona Olivier Bernardi – p.5/37 Example Hard particules in Ω = {1, . . . , q}. λ = 1 and φ(x, y) = +∞ if x = y 0 otherwise. CRM, Barcelona OOlivier Bernardi – p.6/37 Example Hard particules in Ω = {1, . . . , q}. λ = 1 and φ(x, y) = +∞ if x = y 0 otherwise. The partition function : Y X 1 φ(xi , xj ) Z(Ω, T ) ≡ exp − n! Ωn i<j kT CRM, Barcelona OOlivier Bernardi – p.6/37 Example Hard particules in Ω = {1, . . . , q}. λ = 1 and φ(x, y) = +∞ if x = y 0 otherwise. The partition function : Y X 1 φ(xi , xj ) q Z(Ω, T ) ≡ exp − = n! Ωn i<j kT n CRM, Barcelona Olivier Bernardi – p.6/37 Mayer’s Idea (1940) φ(xi , xj ) exp − kT CRM, Barcelona = 1 + f (xi , xj ). OOlivier Bernardi – p.7/37 Mayer’s Idea (1940) φ(xi , xj ) exp − kT i<j Y CRM, Barcelona = Y i<j 1+f (xi , xj ) = X Y f (xi , xj ). G⊆Kn (i,j)∈G OOlivier Bernardi – p.7/37 Mayer’s Idea (1940) φ(xi , xj ) exp − kT i<j Y = Y 1+f (xi , xj ) = i<j X Y f (xi , xj ). G⊆Kn (i,j)∈G ⇒ Partition function can be written as a sum over graphs : ZZ Y 1 φ(xi , xj ) Z(Ω, T, n) ≡ n exp − dx1 ..dxn λ n! Ωn i<j kT 1 X = n W (G), λ n! G⊆K n where W (G) = ZZ Y f (xi , xj )dx1 ..dxn Ωn (i,j)∈G is the Mayer’s weight of G. CRM, Barcelona Olivier Bernardi – p.7/37 For those familiar with the Tutte Polynomial Mayer’s tranformation is the analogue (for general partition function) of the correspondence Partition function of the Potts model (coloring expansion) ⇐⇒ Tutte polynomial (subgraph expansion) [Fortuin & Kasteleyn 72] CRM, Barcelona Olivier Bernardi – p.8/37 Example Hard particules in Ω = {1, . . . , q}. φ(x, y) = +∞ if x = y 0 otherwise. CRM, Barcelona f (x, y) = −1 if x = y 0 otherwise. OOlivier Bernardi – p.9/37 Example Hard particules in Ω = {1, . . . , q}. φ(x, y) = +∞ if x = y 0 otherwise. f (x, y) = −1 if x = y 0 otherwise. Mayer’s weight of G : X Y e(G)q c(G) W (G) = f (xi , xj ) = (−1) . Ωn (i,j)∈G CRM, Barcelona OOlivier Bernardi – p.9/37 Example Hard particules in Ω = {1, . . . , q}. φ(x, y) = +∞ if x = y 0 otherwise. f (x, y) = −1 if x = y 0 otherwise. Mayer’s weight of G : X Y e(G)q c(G) W (G) = f (xi , xj ) = (−1) . Ωn (i,j)∈G Mayer’s correspondence X W (G) = n!Z(Ω, n) shows : G⊆Kn X G⊆Kn CRM, Barcelona (−1)e(G) q c(G) q = n! = q(q − 1) . . . (q − n + 1). n Olivier Bernardi – p.9/37 Allowing any number of particules The grand canonical partition function is X Zgr (Ω, T, z) = Z(Ω, T, n)λn z n . n In terms of Mayer’s weights : Zgr (Ω, T, z) = X n CRM, Barcelona ! X W (G)z |G| 1 X n n W (G) λ z = . n λ n! G⊆K |G|! G n Olivier Bernardi – p.10/37 Pressure The pressure of the system is given by kT P (Ω, T, z) = log (Zgr (Ω, T, z)) . |Ω| CRM, Barcelona OOlivier Bernardi – p.11/37 Pressure The pressure of the system is given by kT P (Ω, T, z) = log (Zgr (Ω, T, z)) . |Ω| Since Mayers weights are multiplicative kT kT P (Ω, T, z) = log (Zgr (Ω, T, z)) = |Ω| |Ω| CRM, Barcelona G X connected W (G)z |G| . |G|! Olivier Bernardi – p.11/37 Example Hard particules in Ω = {1, . . . , q}. Grand canonical partition function : X X q Zgr (Ω, T, z) = Z(Ω, T, n)z n = z n = (1 + z)q . n n n CRM, Barcelona OOlivier Bernardi – p.12/37 Example Hard particules in Ω = {1, . . . , q}. Grand canonical partition function : X X q Zgr (Ω, T, z) = Z(Ω, T, n)z n = z n = (1 + z)q . n n n Pressure : kT P (Ω, T, z) = log (Zgr (Ω, T, z)) = kT log(1 + z). |Ω| CRM, Barcelona Olivier Bernardi – p.12/37 Example Mayer’s weights : W (G) = (−1)e(G) q c(G) . Pressure : kT P (Ω, T, z) = |Ω| CRM, Barcelona G X connected W (G)z |G| = kT |G|! G X connected (−1)e(G) z |G| . |G|! OOlivier Bernardi – p.13/37 Example Mayer’s weights : W (G) = (−1)e(G) q c(G) . Pressure : kT P (Ω, T, z) = |Ω| G X connected W (G)z |G| = kT |G|! G X connected (−1)e(G) z |G| . |G|! Comparing the two expressions of the pressure yields : G CRM, Barcelona X connected (−1)e(G) z |G| = log(1 + z). |G|! OOlivier Bernardi – p.13/37 Example Mayer’s weights : W (G) = (−1)e(G) q c(G) . Pressure : kT P (Ω, T, z) = |Ω| G X connected W (G)z |G| = kT |G|! G X connected (−1)e(G) z |G| . |G|! Comparing the two expressions of the pressure yields : G X connected In other words : G⊆Kn CRM, Barcelona (−1)e(G) z |G| = log(1 + z). |G|! X (−1)e(G) = (−1)n−1 (n − 1)!. connected Olivier Bernardi – p.13/37 How did we get there ? Mayer Z(Ω, T, z) X W (G)z |G| |G|! G log log A CRM, Barcelona = B Olivier Bernardi – p.14/37 A killing involution G⊆Kn CRM, Barcelona X (−1)e(G) = (−1)n−1 (n − 1)! connected OOlivier Bernardi – p.15/37 A killing involution G⊆Kn X (−1)e(G) = (−1)n−1 (n − 1)! connected We define an involution Φ on the set of connected graphs : - Order the edges of Kn lexicographicaly. - Define E ∗ (G) = {e = (i, j) / i and j are connected by G>e }, Φ(G) = G G ⊕ min(E ∗ (G)) CRM, Barcelona if E ∗ (G) =∅ otherwise. OOlivier Bernardi – p.15/37 A killing involution G⊆Kn X (−1)e(G) = (−1)n−1 (n − 1)! connected We define an involution Φ on the set of connected graphs : - Order the edges of Kn lexicographicaly. - Define E ∗ (G) = {e = (i, j) / i and j are connected by G>e }, Φ(G) = G G ⊕ min(E ∗ (G)) if E ∗ (G) =∅ otherwise. Prop [B.] : The only remaining graphs are the increasing spanning trees. (Known to be in bijection with the permutations of {1, .., n − 1}.) CRM, Barcelona Olivier Bernardi – p.15/37 Increasing trees CRM, Barcelona 4 3 4 3 4 3 1 2 1 2 1 2 4 3 4 3 4 3 1 2 1 2 1 2 Olivier Bernardi – p.16/37 For those familiar with the Tutte Polynomial The sum of the Mayer’s weight correspond to the evaluations of TKn (1, 0). This is the number of internal spanning trees. Subgraph expansion ⇐⇒ Spanning tree expansion CRM, Barcelona Olivier Bernardi – p.17/37 Hard-core continuum gas CRM, Barcelona Olivier Bernardi – p.18/37 Hard-core continuum gas Hard particules in Ω = [0, q]. Ω x2 φ(x, y) = +∞ if |x − y| < 1 0 otherwise. CRM, Barcelona x1 x3 f (x, y) = −1 if |x − y| < 1 0 otherwise. OOlivier Bernardi – p.19/37 Hard-core continuum gas Hard particules in Ω = [0, q]. x2 Ω x1 x3 φ(x, y) = +∞ if |x − y| < 1 0 otherwise. W (G) = ZZ Y f (xi , xj )dx1 ..dxn . Ωn (i,j)∈G kT P (Ω, T, z) = |Ω| CRM, Barcelona f (x, y) = −1 if |x − y| < 1 0 otherwise. G X connected W (G)z |G| . |G|! Olivier Bernardi – p.19/37 Thermodynamical limit (|Ω| → ∞) x2 x1 x3 P (T, z) ≡ lim P (Ω, T, z) = kT |Ω|→∞ G X connected W∞ (G)z |G| |G|! where, WΩ (G) = W∞ (G) ≡ lim |Ω|→∞ |Ω| CRM, Barcelona ZZ n−1 ; Y f (xi , xj )dx2 ..dxn . x1 =0 (i,j)∈G Olivier Bernardi – p.20/37 Mayer’s weight for the hard-core gas f (x, y) = −1 if |x − y| < 1 0 otherwise. W (G) = CRM, Barcelona ZZ n−1 ; Y f (xi , xj )dx2 ..dxn . x1 =0 (i,j)∈G OOlivier Bernardi – p.21/37 Mayer’s weight for the hard-core gas f (x, y) = −1 if |x − y| < 1 0 otherwise. W (G) = ZZ n−1 ; Y x1 =0 (i,j)∈G P (T, z) = kT G CRM, Barcelona f (xi , xj )dx2 ..dxn . X connected W (G)z |G| |G|! Olivier Bernardi – p.21/37 Mayer’s diagram for the hard-core gas Mayer Z(T, z) X W (G)z |G| G log (−1) log n−1 n−1 n Cayley trees ! ? CRM, Barcelona |G|! = P G⊆Kn W∞ (G) [Labelle, Leroux, Ducharme : SLC 54] Olivier Bernardi – p.22/37 Slicing W (G) [Lass] W (G) = ZZ n−1 ; f (xi , xj )dx2 ..dxn , x1 =0 (i,j)∈G e(G) × Volume(ΠG ), \ is the polytope |xi − xj | ≤ 1. = (−1) where ΠG ⊂ Rn−1 Y (i,j)∈G CRM, Barcelona OOlivier Bernardi – p.23/37 Slicing W (G) [Lass] W (G) = ZZ n−1 ; f (xi , xj )dx2 ..dxn , x1 =0 (i,j)∈G e(G) × Volume(ΠG ), \ is the polytope |xi − xj | ≤ 1. = (−1) where ΠG ⊂ Rn−1 Y (i,j)∈G Example : G: x1 CRM, Barcelona x3 ΠG : x3 x2 x2 OOlivier Bernardi – p.23/37 Slicing W (G) [Lass] W (G) = ZZ n−1 ; f (xi , xj )dx2 ..dxn , x1 =0 (i,j)∈G e(G) × Volume(ΠG ), \ is the polytope |xi − xj | ≤ 1. = (−1) where ΠG ⊂ Rn−1 Y (i,j)∈G Example : G: x1 CRM, Barcelona x3 ΠG : x3 x2 x2 Olivier Bernardi – p.23/37 Slicing W (G) [Lass] Fractional representation xi = h(xi ) + (xi ). 3 2 h(xi ) = 1 xi 0 −1 =0 CRM, Barcelona (xi ) = 1 OOlivier Bernardi – p.24/37 Slicing W (G) [Lass] Fractional representation xi = h(xi ) + (xi ). |xi − xj | < 1 ⇐⇒ 3 xj 2 1 xi xj xi 0 −1 CRM, Barcelona OOlivier Bernardi – p.24/37 Slicing W (G) [Lass] Fractional representation xi = h(xi ) + (xi ). Prop [Lass] : (x2 , .., xn ) ∈ ΠG ? only depends on the integer parts h(x2 ), .., h(xn ) and the order of the fractional parts (x2 ), .., (xn ). CRM, Barcelona OOlivier Bernardi – p.24/37 Slicing W (G) [Lass] Fractional representation xi = h(xi ) + (xi ). Prop [Lass] : (x2 , .., xn ) ∈ ΠG ? only depends on the integer parts h(x2 ), .., h(xn ) and the order of the fractional parts (x2 ), .., (xn ). h1 h2 h3 h4 h5 h6 =0 =1 =0 =2 = −1 =0 0 = 1 < 4 < 6 < 2 < 5 < 3 CRM, Barcelona 3 2 1 x4 x2 x3 0 x x6 1 x5 −1 OOlivier Bernardi – p.24/37 Slicing W (G) [Lass] Fractional representation xi = h(xi ) + (xi ). Prop [Lass] : (x2 , .., xn ) ∈ ΠG ? only depends on the integer parts h(x2 ), .., h(xn ) and the order of the fractional parts (x2 ), .., (xn ). 3 x6 x1 x5 2 x4 x2 x3 x2 1 0 −1 CRM, Barcelona x4 x6 x1 x3 x5 OOlivier Bernardi – p.24/37 Slicing W (G) [Lass] Fractional representation xi = h(xi ) + (xi ). Prop [Lass] : (x2 , .., xn ) ∈ ΠG ? only depends on the integer parts h(x2 ), .., h(xn ) and the order of the fractional parts (x2 ), .., (xn ). Each subpolytope ∆ defined by h2 , .., hn and an order on 1 (x2 ), .., (xn ) has volume . (n − 1)! CRM, Barcelona Olivier Bernardi – p.24/37 Counting labelled schemes x3 ΠG : x3 G: x1 x2 x2 x1 x1 x2 x3 x3 x2 x1 x2 x1 x3 x1 x2 x3 x1 x2 x3 x2 (−1)e(G) Each labelled scheme has weight . (n − 1)! CRM, Barcelona Olivier Bernardi – p.25/37 Rearanging the sum X W (G) = X (−1)e(G) #{S labelled scheme containing G} (n − 1)! G⊆Kn G⊆Kn connected connected CRM, Barcelona OOlivier Bernardi – p.26/37 Rearanging the sum X W (G) = X (−1)e(G) #{S labelled scheme containing G} (n − 1)! G⊆Kn G⊆Kn connected connected = S CRM, Barcelona X labelled scheme 1 (n − 1)! G X (−1)e(G) contained in S OOlivier Bernardi – p.26/37 Rearanging the sum X X W (G) = (−1)e(G) #{S labelled scheme containing G} (n − 1)! G⊆Kn G⊆Kn connected connected = S labelled scheme = S CRM, Barcelona X X scheme G 1 (n − 1)! X G X (−1)e(G) contained in S (−1)e(G) contained in S Olivier Bernardi – p.26/37 Rearranging the sum 3 2 1 0 −1 CRM, Barcelona G X (−1)e(G) contained in S Olivier Bernardi – p.27/37 A killing involution We define an involution Φ on the set of connected graphs contained in S : - Order the edges of Kn lexicographicaly. - Define E ∗ (G) = {e = (i, j) / i and j are connected by G>e }, Φ(G) = G G ⊕ min(E ∗ (G)) CRM, Barcelona if E ∗ (G) =∅ otherwise. OOlivier Bernardi – p.28/37 A killing involution We define an involution Φ on the set of connected graphs contained in S : - Order the edges of Kn lexicographicaly. - Define E ∗ (G) = {e = (i, j) / i and j are connected by G>e }, Φ(G) = G G ⊕ min(E ∗ (G)) if E ∗ (G) =∅ otherwise. Proposition [B.] : The only remaining graphs are the increasing spanning trees. Corrolary : X G CRM, Barcelona (−1)e(G) = (−1)n−1 #{increasing tree on S}. contained in S Olivier Bernardi – p.28/37 A killing involution CRM, Barcelona OOlivier Bernardi – p.29/37 A killing involution CRM, Barcelona Olivier Bernardi – p.29/37 Bijection with Cayley trees Theorem [B.] : S [ {increasing tree} are in bijection with scheme rooted Cayley trees. CRM, Barcelona OOlivier Bernardi – p.30/37 Bijection with Cayley trees Theorem [B.] : S [ {increasing tree} are in bijection with scheme rooted Cayley trees. CRM, Barcelona OOlivier Bernardi – p.30/37 Bijection with Cayley trees Theorem [B.] : S [ {increasing tree} are in bijection with scheme rooted Cayley trees. 4 6 2 1 9 7 3 8 5 CRM, Barcelona OOlivier Bernardi – p.30/37 Bijection with Cayley trees Theorem [B.] : S [ {increasing tree} are in bijection with scheme rooted Cayley trees. 4 6 2 1 9 7 3 8 5 CRM, Barcelona Olivier Bernardi – p.30/37 Bijection with Cayley trees Corollary [B.] : X W (G) = G⊆Kn connected S X X (−1)e(G) scheme G contained in S = (−1) n−1 S X #{increasing tree on S} scheme = (−1)n−1 nn−1 . CRM, Barcelona Olivier Bernardi – p.31/37 Bijection with Cayley trees 1 2 1 3 3 3 2 1 2 1 3 1 3 CRM, Barcelona 2 3 2 3 2 1 2 1 3 1 2 3 2 1 Olivier Bernardi – p.32/37 Concluding remarks CRM, Barcelona Olivier Bernardi – p.33/37 Mayer’s transformation Producing graph weights Mayer Z(Ω, T, z) X W (G)z |G| G CRM, Barcelona |G|! OOlivier Bernardi – p.34/37 Mayer’s transformation Producing graph weights Producing nasty identities Mayer Z(Ω, T, z) X W (G)z |G| |G|! G log log A CRM, Barcelona = B Olivier Bernardi – p.34/37 Discrete hard-core gas (colorings) Mayer Z(Ω, T, z) X W (G)z |G| G log (−1)n−1 (n − 1)! |G|! log = X (−1)e(G) G⊆Kn connected CRM, Barcelona OOlivier Bernardi – p.35/37 Discrete hard-core gas (colorings) Potts model ⇐⇒ Subgraph expansion Mayer Z(Ω, T, z) X W (G)z |G| G log (−1)n−1 (n − 1)! |G|! log = X (−1)e(G) G⊆Kn Spanning tree expansion CRM, Barcelona ⇐⇒ connected Subgraph expansion Olivier Bernardi – p.35/37 Continuous hard-core gas Mayer Z(T, z) X W (G)z |G| G log (−1) CRM, Barcelona |G|! log n−1 n−1 n = P G⊆Kn W∞ (G) Olivier Bernardi – p.36/37 Thanks. CRM, Barcelona Olivier Bernardi – p.37/37