Abstract
Our work deals with symmetric rational functions and probabilistic models based on the fully inhomogeneous six vertex (ice type) model satisfying the free fermion condition. Two families of symmetric rational functions \(F_\lambda ,G_\lambda \) are defined as certain partition functions of the six vertex model, with variables corresponding to row rapidities, and the labeling signatures \(\lambda =(\lambda _1\ge \ldots \ge \lambda _N)\in {\mathbb {Z}}^N\) encoding boundary conditions. These symmetric functions generalize Schur symmetric polynomials, as well as some of their variations, such as factorial and supersymmetric Schur polynomials. Cauchy type summation identities for \(F_\lambda ,G_\lambda \) and their skew counterparts follow from the Yang–Baxter equation. Using algebraic Bethe Ansatz, we obtain a double alternant type formula for \(F_\lambda \) and a Sergeev–Pragacz type formula for \(G_\lambda \). In the spirit of the theory of Schur processes, we define probability measures on sequences of signatures with probability weights proportional to products of our symmetric functions. We show that these measures can be viewed as determinantal point processes, and we express their correlation kernels in a double contour integral form. We present two proofs: The first is a direct computation of Eynard–Mehta type, and the second uses non-standard, inhomogeneous versions of fermionic operators in a Fock space coming from the algebraic Bethe Ansatz for the six vertex model. We also interpret our determinantal processes as random domino tilings of a half-strip with inhomogeneous domino weights. In the bulk, we show that the lattice asymptotic behavior of such domino tilings is described by a new determinantal point process on \({\mathbb {Z}}^{2}\), which can be viewed as an doubly-inhomogeneous generalization of the extended discrete sine process.
Similar content being viewed by others
Notes
The term “bulk” refers to the parts of the system where the space can be rescaled to form growing regions with unit particle density.
All square roots involved in identities in this proposition and throughout the section are always squared in the action of the operators, so we do not need to specify the branches.
References
Aggarwal, A., Borodin, A., Wheeler, M.: Colored Fermionic Vertex Models and Symmetric Functions, arXiv preprint (2021). arXiv:2101.01605 [math.CO]
Aggarwal, A.: Universality for Lozenge Tiling Local Statistics, arXiv preprint (2019). arXiv:1907.09991 [math.PR]
Anderson, G.W., Guionnet, A., Zeitouni, O.: An Introduction to Random Matrices. Cambridge University Press, Cambridge (2010)
Assiotis, T.: Determinantal structures in space inhomogeneous dynamics on interlacing arrays. Ann. Inst. H. Poincaré 21, 909–940 (2020). arXiv:1910.09500 [math.PR]
Baxter, R.: Exactly Solved Models in Statistical Mechanics. Academic Press, London (1989)
Betea, D., Bouttier, J.: The periodic Schur process and free fermions at finite temperature. Math. Phys. Anal. Geom. 22(1), 3 (2019). arXiv:1807.09022 [math-ph]
Boutillier, C., Bouttier, J., Chapuy, G., Corteel, S., Ramassamy, S.: Dimers on rail yard graphs. Ann. Inst. Henri Poincaré D 4(4), 479–539 (2017). arXiv:1504.05176 [math-ph]
Brubaker, B., Bump, D., Friedberg, S.: Schur polynomials and the Yang–Baxter equation. Commun. Math. Phys. 308(2), 281 (2011)
Borodin, A., Corwin, I.: Macdonald processes. Probab. Theory Relat. Fields 158, 225–400 (2014). arXiv:1111.4408 [math.PR]
Bouttier, J., Chapuy, G., Corteel, S.: From Aztec diamonds to pyramids: steep tilings. Trans. AMS 369(8), 5921–5959 (2017). arXiv:1407.0665 [math.CO]
Berggren, T.: Domino tilings of the Aztec diamond with doubly periodic weightings. Ann. Probab. 49(4), 1965–2011 (2021). arXiv:1911.01250 [math.PR]
Borodin, A., Ferrari, P.: Anisotropic growth of random surfaces in 2+1 dimensions. Commun. Math. Phys. 325, 603–684 (2014). arXiv:0804.3035 [math-ph]
Bump, D., McNamara, P., Nakasuji, M.: Factorial Schur functions and the Yang–Baxter equation. Rikkyo-daigaku-sugaku-zasshi 63(1–2), 23–45 (2014). arXiv:1108.3087 [math.CO]
Borodin, A., Olshanski, G.: Distributions on partitions, point processes, and the hypergeometric kernel. Commun. Math. Phys. 211(2), 335–358 (2000). arXiv:math/9904010 [math.RT]
Borodin, A., Okounkov, A., Olshanski, G.: Asymptotics of Plancherel measures for symmetric groups. J. AMS 13(3), 481–515 (2000). arXiv:math/9905032 [math.CO]
Borodin, A.: Periodic Schur process and cylindric partitions. Duke J. Math. 140(3), 391–468 (2007). arXiv:math/0601019 [math.CO]
Borodin, A.: Determinantal Point Processes, Oxford Handbook of Random Matrix Theory (2011). arXiv:0911.1153 [math.PR]
Borodin, A., Peche, S.: Airy kernel with two sets of parameters in directed percolation and random matrix theory. J. Stat. Phys. 132(2), 275–290 (2008). arXiv:0712.1086v3 [math-ph]
Borodin, A., Petrov, L.: Higher spin six vertex model and symmetric rational functions. Selecta Math. 24(2), 751–874 (2018). arXiv:1601.05770 [math.PR]
Borodin, A., Rains, E.M.: Eynard–Mehta theorem, Schur process, and their Pfaffian analogs. J. Stat. Phys. 121(3), 291–317 (2005). arXiv:math-ph/0409059
Berele, A., Regev, A.: Hook Young diagrams with applications to combinatorics and representations of Lie superalgebras. Adv. Math. 64(2), 118–175 (1987)
Borodin, A., Shlosman, S.: Gibbs ensembles of nonintersecting paths. Commun. Math. Phys. 293(1), 145–170 (2010). arXiv:0804.0564 [math-ph]
Cohn, H., Elkies, N., Propp, J.: Local statistics for random domino tilings of the Aztec diamond. Duke Math. J. 85(1), 117–166 (1996). arXiv:math/0008243 [math.CO]
Charlier, C.: Doubly periodic lozenge tilings of a hexagon and matrix valued orthogonal polynomials. Stud. Appl. Math. 146(1), 3–80 (2021). arXiv:2001.11095 [math-ph]
Chhita, S., Johansson, K.: Domino statistics of the two-periodic Aztec diamond. Adv. Math. 294, 37–149 (2016). arXiv:1410.2385 [math.PR]
Cohn, H., Kenyon, R., Propp, J.: A variational principle for domino tilings. J. AMS 14(2), 297–346 (2001). arXiv:math/0008220 [math.CO]
Decreusefond, L., Flint, I., Privault, Nicolas, Giovanni L.T.: Determinantal point processes. Stochastic Analysis for Poisson Point Processes, pp. 311–342 (2016)
Duits, M., Kuijlaars, A.: The two periodic Aztec diamond and matrix valued orthogonal polynomials. J. Eur. Math. Soc. 23(4), 1075–1131 (2020). arXiv:1712.05636 [math.PR]
Dyson, F.J.: A Brownian motion model for the eigenvalues of a random matrix. J. Math. Phys. 3(6), 1191–1198 (1962)
Elkies, N., Kuperberg, G., Larsen, M., Propp, J.: Alternating-sign matrices and domino tilings. J. Alg. Combin. 1(2–3), 111-132–219-234 (1992)
Eynard, B., Mehta, M.L.: Matrices coupled in a chain: I. Eigenvalue correlations. J. Phys. A 31, 4449–4456 (1998)
Faddeev, L.D.: How algebraic Bethe ansatz works for integrable model (1996). Les-Houches lectures (1996). arXiv:hep-th/9605187
Felderhof, B.U.: Diagonalization of the transfer matrix of the free-fermion model. II. Physica 66(2), 279–297 (1973)
Felderhof, B.U.: Diagonalization of the transfer matrix of the free-fermion model. III. Physica 66(3), 509–526 (1973)
Felderhof, B.U.: Direct diagonalization of the transfer matrix of the zero-field free-fermion model. Physica 65(3), 421–451 (1973)
Fomin, S., Kirillov, A.N.: Grothendieck polynomials and the Yang–Baxter equation. Proceedings of Formal Power Series and Algebraic Combinatorics, pp. 183–190 (1994)
Fomin, S., Kirillov, A.N.: The Yang–Baxter equation, symmetric functions, and Schubert polynomials. Discrete Math. 153(1–3), 123–143 (1996)
Fehér, L., Némethi, A., Rimányi, R.: Equivariant classes of matrix matroid varieties. Comment. Math. Helv. 87(4), 861–889 (2012). arXiv:0812.4871 [math.AG]
Forrester, P.J.: Meet Andréief, Bordeaux 1886, and Andreev, Kharkov 1882–1883. Random Matrices Theory Appl. 8(02), 1930001 (2019). arXiv:1806.10411 [math-ph]
Ferrari, P.L., Spohn, H.: Domino tilings and the six-vertex model at its free-fermion point. J. Phys. A 39(33), 10297 (2006). arXiv:cond-mat/0605406 [cond-mat.stat-mech]
Felder, G., Varchenko, A.: Algebraic Bethe ansatz for the elliptic quantum group \(E_{\tau ,\eta }({\rm sl}_2)\). Nucl. Phys. B 480(1–2), 485–503 (1996). arXiv:q-alg/9605024
Gaudin, M.: Une démonstration simplifiée du théoreme de wick en mécanique statistique. Nucl. Phys. 15, 89–91 (1960)
Gorin, Vadim: Lectures on random lozenge tilings, Cambridge Studies in Advanced Mathematics. Cambridge University Press (2021). https://people.math.wisc.edu/~vadicgor/Random_tilings.pdf
Gleizer, O., Postnikov, A.: Littlewood–Richardson coefficients via Yang–Baxter equation. Int. Math. Res. Not. 2000(14), 741–774 (2000)
Gorin, V., Petrov, L.: Universality of local statistics for noncolliding random walks. Ann. Probab. 47(5), 2686–2753 (2019). arXiv:1608.03243 [math.PR]
Guo, P., Sun, S.: Identities on factorial grothendieck polynomials. Adv. Appl. Math. 111, 101933 (2019). arXiv:1812.04390 [math.CO]
Gunna, A., Scrimshaw, T.: Integrable systems and crystals for edge labeled tableaux, arXiv preprint (2022). arXiv:2202.06004 [math.CO]
Hardt, A.: Lattice Models, Hamiltonian Operators, and Symmetric Functions, arXiv preprint (2021). arXiv:2109.14597 [math.RT]
Hamel, A.M., Goulden, I.P.: Lattice paths and a Sergeev–Pragacz formula for skew supersymmetric functions. Can. J. Math. 47(2), 364–382 (1995)
Hough, J.B., Krishnapur, M., Peres, Y., Virág, B.: Determinantal processes and independence. Probab. Surv. 3, 206–229 (2006). arXiv:math/0503110 [math.PR]
Ikeda, T., Naruse, H.: Excited Young diagrams and equivariant Schubert calculus. Trans. AMS 361(10), 5193–5221 (2009). arXiv:math/0703637 [math.AG]
Johansson, K.: Non-intersecting paths, random tilings and random matrices. Probab. Theory Relat. Fields 123(2), 225–280 (2002). arXiv:math/0011250 [math.PR]
Johansson, K.: The arctic circle boundary and the Airy process. Ann. Probab. 33(1), 1–30 (2005). arXiv:math/0306216 [math.PR]
Johansson, K.: Random matrices and determinantal processes (2005). arXiv:math-ph/0510038
Kac, Victor G.: Infinite-Dimensional Lie Algebras, 3rd edn. Cambridge University Press, Cambridge (1990) MR1104219 (92k:17038)
Korepin, V., Bogoliubov, N., Izergin, A.: Quantum Inverse Scattering Method and Correlation Functions. Cambridge University Press, Cambridge (1993)
Kenyon, R.: Dominos and the Gaussian free field. Ann. Probab. 29(3), 1128–1137 (2001). arXiv:math-ph/0002027
Kenyon, R.: Lectures on dimers (2009). arXiv:0910.3129 [math.PR]
Kitanine, N., Maillet, J.M., Slavnov, N.A., Terras, V.: Spin–spin correlation functions of the XXZ-12 Heisenberg chain in a magnetic field. Nucl. Phys. B 641(3), 487–518 (2002). arXiv:hep-th/0201045
Kenyon, R., Okounkov, A.: Limit shapes and the complex Burgers equation. Acta Math. 199(2), 263–302 (2007). arXiv:math-ph/0507007
König, W.: Orthogonal polynomial ensembles in probability theory. Probab. Surv. 2, 385–447 (2005). arXiv:math/0403090 [math.PR]
König, W., O’Connell, N., Roch, S.: Non-colliding random walks, tandem queues, and discrete orthogonal polynomial ensembles. Electron. J. Probab. 7(5), 1–24 (2002)
Korff, C.: Cylindric versions of specialised Macdonald functions and a deformed Verlinde algebra. Commun. Math. Phys. 318(1), 173–246 (2013). arXiv:1110.6356 [math-ph]
Korff, C.: Cylindric Hecke characters and Gromov–Witten invariants via the asymmetric six-vertex model. Commun. Math. Phys. 381(2), 591–640 (2021). arXiv:1906.02565 [math-ph]
Kenyon, R., Okounkov, A., Sheffield, S.: Dimers and amoebae. Ann. Math. 163, 1019–1056 (2006). arXiv:math-ph/0311005
Kirillov, A.N., Reshetikhin, N.Y.: The Bethe ansatz and the combinatorics of Young tableaux. J. Sov. Math. 41(2), 925–955 (1988)
Kulesza, A., Taskar, B.: Determinantal point processes for machine learning. Found. Trends Mach. Learn. 5(2–3), 123–286 (2012). arXiv:1207.6083 [stat.ML]
Lascoux, A.: The 6 vertex model and Schubert polynomials. SIGMA 3, 029 (2007). arXiv:math/0610719 [math.CO]
Lascoux, A., Leclerc, B., Thibon, J.-Y.: Flag varieties and the Yang–Baxter equation. Lett. Math. Phys. 40(1), 75–90 (1997)
Lyons, R.: Determinantal probability measures. Publ. IHES 98, 167–212 (2003). arXiv:math/0204325 [math.PR]
Macchi, O.: The coincidence approach to stochastic point processes. Adv. Appl. Probab. 7(1), 83–122 (1975)
Macdonald, I.G.: Schur functions: theme and variations. Sém. Lothar. Combin. 28, 5–39 (1992)
Macdonald, I.G.: Symmetric Functions and Hall Polynomials, 2nd edn. Oxford University Press, Oxford (1995)
McNamara, P.: Factorial Schur functions via the six vertex model, arXiv preprint (2009). arXiv:0910.5288 [math.CO]
Mehta, M.L., Gaudin, M.: On the density of eigenvalues of a random matrix. Nucl. Phys. 18, 420–427 (1960)
Mkrtchyan, S.: Plane partitions with 2-periodic weights. Lett. Math. Phys. 104(9), 1053–1078 (2014). arXiv:1309.4825 [math.PR]
Mkrtchyan, S.: Plane partitions with 2-periodic weights. Lett. Math. Phys. 104(9), 1053–1078 (2014). arXiv:1309.4825 [math.PR]
Molev, A.: Comultiplication rules for the double Schur functions and Cauchy identities. Electron. J. Comb. R13 (2009). arXiv:0807.2127 [math.CO]
Motegi, K.: Izergin–Korepin analysis on the projected wavefunctions of the generalized free-fermion model. Adv. Math. Phys. 2017, 7563781 (2017). arXiv:1704.03575 [math-ph]
Motegi, K.: Integrability approach to Fehér-Némethi-Rimányi-Guo-Sun type identities for factorial Grothendieck polynomials. Nucl. Phys. B 954, 114998 (2020). arXiv:1909.02278 [math.CO]
Morales, A.H., Pak, I., Panova, G.: Hook formulas for skew shapes II. Combinatorial proofs and enumerative applications. SIAM J. Discrete Math. 31(3), 1953–1989 (2017). arXiv:1610.04744 [math.CO]
Morales, A., Pak, I., Panova, G.: Hook formulas for skew shapes III. Multivariate and product formulas. Alg. Combin. 2(5), 815–861 (2019). arXiv:1707.00931 [math.CO]
Moens, E.M., Van der Jeugt, J.: A determinantal formula for supersymmetric Schur polynomials. J. Alg. Combin. 17(3), 283–307 (2003)
Nagao, T., Forrester, P.J.: Multilevel dynamical correlation functions for Dyson’s Brownian motion model of random matrices. Phys. Lett. A 247(1–2), 42–46 (1998)
Nakagawa, J., Noumi, M., Shirakawa, M., Yamada, Y.: Tableau representation for Macdonald’s ninth variation of Schur functions. Phys. Combin. (2001). https://doi.org/10.1142/9789812810007_0008
Okounkov, A.: Infinite wedge and random partitions. Selecta Math. 7(1), 57–81 (2001). arXiv:math/9907127 [math.RT]
Okounkov, A.: Symmetric functions and random partitions, Symmetric functions 2001: Surveys of developments and perspectives (2002). arXiv:math/0309074 [math.CO]
Olshanski, G.: Interpolation Macdonald polynomials and Cauchy-type identities. J. Combin. Theory A 162, 65–117 (2019). arXiv:1712.08018 [math.CO]
Oota, T.: Quantum projectors and local operators in lattice integrable models. J. Phys. A 37(2), 441 (2003). arXiv:hep-th/0304205
Okounkov, A., Reshetikhin, N.: Correlation function of Schur process with application to local geometry of a random 3-dimensional Young diagram. J. AMS 16(3), 581–603 (2003). arXiv:math/0107056 [math.CO]
Pauling, L.: The structure and entropy of ice and of other crystals with some randomness of atomic arrangement. J. Am. Chem. Soc. 57(12), 2680–2684 (1935)
Petrov, L.: Asymptotics of uniformly random lozenge tilings of polygons. Gaussian free field. Ann. Probab. 43(1), 1–43 (2015), available at 1206.5123. arXiv:1206.5123 [math.PR]
Pak, I., Petrov, F.: Hidden symmetries of weighted lozenge tilings. Electron. J. Combin. 27(3), 3–44 (2020). arXiv:2003.14236 [math.CO]
Reshetikhin, N.: Lectures on the integrability of the 6-vertex model, Exact Methods in Low-dimensional Statistical Physics and Quantum Computing, pp. 197–266 (2010). arXiv:1010.5031 [math-ph]
Sheffield, S.: Random surfaces, Astérisque 304 (2005). arXiv:math/0304049 [math.PR]
Soshnikov, A.: Determinantal random point fields. Russ. Math. Surv. 55(5), 923–975 (2000). arXiv:math/0002099 [math.PR]
Tsilevich, N.: Quantum inverse scattering method for the q-boson model and symmetric functions. Funct. Anal. Appl. 40(3), 207–217 (2006). arXiv:math-ph/0510073
Wheeler, M., Zinn-Justin, P.: Hall polynomials, inverse Kostka polynomials and puzzles. J. Combin. Theory A 159, 107–163 (2018). arXiv:1603.01815 [math-ph]
Yau, H.-T.: The Wigner–Dyson–Gaudin-Mehta Conjecture. Notices of the international congress of Chinese mathematicians, pp. 10–13 (2013)
Zinn-Justin, P.: Six-vertex model with domain wall boundary conditions and one-matrix model. Phys. Rev. E 62(3), 3411 (2000). arXiv:math-ph/0005008
Zinn-Justin, P.: Littlewood–Richardson coefficients and integrable tilings. Electron. J. Combin. 16(R12), 1 (2009). arXiv:0809.2392 [math-ph]
Zinn-Justin, P.: Six-Vertex. Integrability and Combinatorics, Loop and Tiling models (2009). arXiv:0901.0665 [math-ph]
Acknowledgements
Amol Aggarwal was partially supported by a Clay Research Fellowship. Alexei Borodin was partially supported by the NSF grants DMS-1664619, DMS-1853981, and the Simons Investigator program. Leonid Petrov was partially supported by the NSF grant DMS-1664617, and the Simons Collaboration Grant for Mathematicians 709055. Michael Wheeler was partially supported by an Australian Research Council Future Fellowship, grant FT200100981. This material is based upon work supported by the National Science Foundation under Grant No. DMS-1928930 while Aggarwal and Petrov participated in program hosted by the Mathematical Sciences Research institute in Berkeley, California, during the Fall 2021 semester. We are very grateful to the anonymous referees for numerous helpful remarks.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Part IV Appendix
Formulas for \(F_\lambda \) and \(G_\lambda \)
Here we employ the row operators (defined in Sect. 2.3) to get explicit formulas for the partition functions \(F_\lambda \) and \(G_\lambda \) of the free fermion six vertex model, and thus prove Theorems 3.9 and 3.10. This Appendix accompanies Sect. 3 and employs algebraic Bethe Ansatz type computations. They follow [19, Section 4.5] (but are more involved in the case of \(G_\lambda \)), see also Part VII and in particular Appendix VII.2 of [56].
1.1 Proof of Theorem 3.9
1.1.1 Recalling the notation
Throughout this subsection we fix a signature \(\lambda =(\lambda _1,\ldots ,\lambda _N \ge 0)\) with N parts, and sequences
Recall (Definition 3.3) that the function \(F_\lambda ({\textbf{x}};{\textbf{y}};{\textbf{r}};{\textbf{s}})\) is the partition function of the free fermion six vertex model with weights \({\widehat{W}}\) (2.4) and with boundary conditions determined by \(\lambda \).
In this subsection we prove Theorem 3.9 stating that \(F_\lambda \) is given by the determinantal expression (3.12) involving the functions \(\varphi _k(x)\) (3.11). For convenience, let us explicitly reproduce the desired formula here:
For the proof we will need the row operators \({\widehat{A}},{\widehat{B}},{\widehat{C}},{\widehat{D}}\) defined by (2.22)–(2.23). These operators are built from the weights \({\widehat{W}}\), depend on two numbers x, r and the sequences \({\textbf{y}}, {\textbf{s}}\), and act (from the right) on tensor products of two-dimensional spaces \(V^{(k)}=\mathop {\textrm{span}}\{ e_0^{(k)},e_1^{(k)} \} \simeq {\mathbb {C}}^2\). To the signature \(\lambda \) we associate the element \(e_{{\mathcal {S}}(\lambda )}\) in the (formal) infinite tensor product \(V^{(1)}\otimes V^{(2)}\otimes \ldots \), where we take \(e^{(k)}_1\) in the k-th place if and only if \(k\in {\mathcal {S}}(\lambda )\) and \(e_0^{(k)}\) otherwise, see Sect. 3.1. For example, the empty signature \(\varnothing \) (which has 0 parts) corresponds to \(e_{\varnothing }=e_0^{(1)}\otimes e_0^{(2)}\otimes \ldots \).
By Proposition 3.4, \(F_\lambda ({\textbf{x}};{\textbf{y}};{\textbf{r}};{\textbf{s}})\) is the coefficient of \(e_{{\mathcal {S}}(\lambda )}\) in \(e_{\varnothing }{\widehat{B}}(x_N,r_N)\ldots {\widehat{B}}(x_1,r_1) \), and for the proof of Theorem 3.9 we proceed to evaluate this coefficient. One of our main tools is the Yang–Baxter equation stated as a family of commutation relations between the operators (see Proposition 2.5).
1.1.2 Action on a tensor product of two spaces
The crucial part of the argument is to consider the action of \({\widehat{B}}(x_N,r_N)\ldots {\widehat{B}}(x_1,r_1)\) on a tensor product of two spaces, \(V_1\otimes V_2\). Using the second identity from (2.23), namely, \((v_1 \otimes v_2) {\widehat{B}} = v_1 {\widehat{D}} \otimes v_2 {\widehat{B}} + v_1 {\widehat{B}} \otimes v_2 {\widehat{A}}\), we see that
where
Now, using the commutation relations (2.26)–(2.27) from Proposition 2.5, we move all the operators \({{\widehat{B}}}\) to the right in both \(X_{\mathcal {I}}\) and \(Y_{\mathcal {I}}\), which allows to rewrite (A.2) as
for some rational functions \(c_{I; I'} ({\textbf{x}}; {\textbf{r}})\), where we have denoted \(|I| = k\) and \(|I'| = m\), defined \(J = \left\{ 1,\ldots ,N \right\} {\setminus } I\) and \(J' = \left\{ 1,\ldots ,N \right\} {\setminus } I'\), and ordered the indices such that \(i_\alpha<i_\beta ,i_\alpha '<i_\beta ',j_\alpha <j_\beta \), and \(j_\alpha '<j_\beta '\) for all \(\alpha <\beta \). Here we also employed the commutativity of \({\widehat{A}}\) (2.24) and \({\widehat{D}}\) (2.28). In fact, here one can already see from (2.26)–(2.27) that \(m=N-k\), but we will get this relation (and a stronger relation between the sets \(I,I',J,J'\)) in the next Lemma A.2.
Remark A.1
Let us make an important observation about the coefficients \(c_{I;I'}({\textbf{x}};{\textbf{r}})\). Namely, these coefficients are computed using only the commutation relations for the operators \({\widehat{A}},{\widehat{B}},{\widehat{C}},{\widehat{D}}\), and we argue that the \(c_{I;I'}({\textbf{x}};{\textbf{r}})\)’s do not depend on the order of applying the commutation relations. This property is based on the fact that for generic parameters (x, r), there exists a representation of \(\begin{bmatrix} {\widehat{A}}(x,r)&{}{\widehat{B}}(x,r)\\ {\widehat{C}}(x,r)&{}{\widehat{D}}(x,r) \end{bmatrix}\) subject to the same commutation relations, and a highest weight vector (annihilated by \({\widehat{C}}\) and an eigenfunctions of \({\widehat{A}},{\widehat{D}}\)) \({\textsf{v}}_0\) in that representation, such that vectors \(\Big (\prod _{k\in {\mathcal {K}}}{\widehat{B}}(x_k,r_k)\Big ){\textsf{v}}_0\), with \({\mathcal {K}}\) ranging over all subsets of \(\{1,2,\ldots ,N\}\), are linearly independent. This fact is a corollary of [41, Lemma 14]: our operators are based on the free fermion six vertex weights, and the cited paper deals with more general eight vertex case.
Therefore, if we apply the commutation relations in two ways and get different coefficients \(c_{I;I'}({\textbf{x}};{\textbf{r}})\) in (A.3), then we can apply these commutation relations in the above highest weight representation, which contradicts the linear independence.
Lemma A.2
We have \(c_{I;I'}({\textbf{x}};{\textbf{r}})=0\) if \(I\cap I'\ne \varnothing \) or \(J\cap J'\ne \varnothing \).
Proof
The two claims with \(I\cap I'\ne \varnothing \) and \(J\cap J'\ne \varnothing \) are analogous, so we only prove the first one.
Suppose \(I\cap I'\ne \varnothing \). Since the operators \({{\widehat{B}}}(x,r)\) commute up to a scalar factor (see (2.25)), we may assume that \(I\cap I'\ni N\) by permuting terms in the left-hand side of (A.2).
Observe that no summand in (A.2) with \(X_N ({\mathcal {I}}; x_N, r_N) = {\widehat{D}} (x_N, r_N)\) (i.e., \(N\in {\mathcal {I}}\)) contributes to a nonzero value of \(c_{I; I'} ({\textbf{x}}, {\textbf{r}}) \). Indeed, in this case the operator \({\widehat{D}}(x_N,r_N)\) is the leftmost term in \(X_{{\mathcal {I}}}\), and thus it does not get involved in the commutation relations of the form (2.26), which means that one cannot obtain \({\widehat{B}}(x_N,r_N)\) from this term. Similarly, no summand in (A.2) with \(Y_N ({\mathcal {I}}; x_N, r_N) = {\widehat{A}} (x_N, r_N)\) (i.e., \(N\notin {\mathcal {I}}\)) contributes to a nonzero value of \(c_{I; I'} ({\textbf{x}}; {\textbf{r}})\).
However, for any \({\mathcal {I}}\subset \left\{ 0,1 \right\} ^{N}\) we either have \(X_N ({\mathcal {I}}; x_N, r_N) = {\widehat{D}} (x_N, r_N)\) or \(Y_N({\mathcal {I}}; x_N, r_N) = {\widehat{A}} (x_N, r_N)\), and so we cannot obtain \({\widehat{B}}(x_N,r_N)\) in both tensor factors. Therefore, terms with \(I\cap I'\ne \varnothing \) are zero. \(\square \)
We see that in (A.3) it must be \(I=J'\) and \(I'=J\), and we may abbreviate \(c_{I;I'}=c_I\). We thus rewrite (A.2)–(A.3) as
We will now evaluate the coefficients \(c_I({\textbf{x}};{\textbf{r}})\). First, set \(I=\left\{ N-k+1,N-k+2,\ldots , N \right\} \). Then the operator
might come from (A.2) only for \({\mathcal {I}}=\{1,2,\ldots ,N-k \}\), in which case
In the first term, we use the commutation relation (2.26) to place the \({\widehat{D}}\) operators on the left and extract the coefficient \(c_{I}({\textbf{x}};{\textbf{r}})\) of (A.5). We have thus established:
Lemma A.3
For \(I=I_k:=\left\{ N-k+1,N-k+2,\ldots ,N \right\} \), the rational function \(c_I\) is equal to
We are now in a position to compute \(c_I({\textbf{x}};{\textbf{r}})\) for arbitrary \(I\subset \left\{ 1,\ldots ,N \right\} \) of size k (where k is also arbitrary) by permuting the \({\widehat{B}}\) operators in the left-hand side of (A.2) thanks to the commutation relation (2.25). For each such I, let \(\sigma \) be a permutation of \(\left\{ 1,\ldots ,N \right\} \) which is increasing on the intervals \(\left\{ 1,\ldots ,N-k \right\} \) and \(\left\{ N-k+1,\ldots ,N \right\} \), and sends \(\left\{ N-k+1,N-k+2,\ldots ,N \right\} \) to I.
Lemma A.4
With the above notation, we have
Proof
The claim follows from the fact that
which in turn holds thanks to (2.25) via induction on the length of the permutation \(\sigma \) (which is the minimal number of elementary transpositions required to represent \(\sigma \) as their product). Then we can further simplify:
which leads to the desired right-hand side of (A.6). The signature of the permutation \(\sigma \) arises by turning \(x_i-x_j\) into \(x_j-x_i\) for each pair \(i\notin I,\,j\in I\) with \(i>j\). \(\square \)
1.1.3 Completing the proof
We are now in a position to prove the determinantal formula (A.1), which finalizes the proof of Theorem 3.9. The goal is to express the coefficient of \(e_{{\mathcal {S}}(\lambda )}\) in \(e_{\varnothing }{\widehat{B}}(x_N,r_N)\ldots {\widehat{B}}(x_1,r_1) \). We are going to repeatedly apply identity (A.4) (with the coefficients \(c_I({\textbf{x}};{\textbf{r}})\) given by (A.6)) to vectors of the form \(e_0^{(m)} \otimes v\), \(m=1,2,\ldots \).
Observe that \(e_0^{(m)}{\widehat{B}}(x,r){\widehat{B}}(x',r')=0\). Therefore, any nonzero summand in (A.4) must have \(|I| \le 1\). Moreover, when \(I=\{i\}\) has one element, any such nonzero contribution to the coefficient of \(e_{{\mathcal {S}}(\lambda )}\) should have \(i\in {\mathcal {S}}(\lambda )=\{\lambda _N+1, \lambda _{N-1}+2, \ldots , \lambda _1+N \}\). Therefore, each step of the repeated application of (A.4) for which we choose \(|I| = 1\) corresponds to a number from 1 to N (indicating which element of \({\mathcal {S}}(\lambda )\) is selected), and these numbers must be distinct. We encode this information by a permutation \(\tau \in {\mathfrak {S}}_N\). Using the facts that
we see that the coefficient of \(e_{{\mathcal {S}}(\lambda )}\) in \(e_{\varnothing }{\widehat{B}}(x_N,r_N)\ldots {\widehat{B}}(x_1,r_1) \) is equal to
Note that the prefactor \(\prod _{1\le i<j\le N}\frac{r_i^{-2}x_i-x_j}{x_i-x_j}\) arises by taking the product of the \(c_{ \{k \} }\)’s over all \(k=1,\ldots ,N\), but in this product for each next term the number N of variables decreases by one. Therefore, we end up with a product over \(i<j\) instead of over all pairs \(i\ne j\). This completes the proof of Theorem 3.9.
1.2 Proof of Theorem 3.10
1.2.1 Recalling the notation
Throughout this subsection we fix \(M,N\ge 1\), a signature \(\lambda =(\lambda _1\ge \ldots \ge \lambda _N \ge 0)\) with N parts, and sequences of complex parameters
Recall from Definition 3.2 the function \(G_\lambda ({\textbf{x}};{\textbf{y}};{\textbf{r}};{\textbf{s}}) =G_{\lambda /0^N}({\textbf{x}};{\textbf{y}};{\textbf{r}};{\textbf{s}})\) which is the partition function of the free fermion six vertex model with weights W (2.3) and with boundary conditions determined by \(\lambda \).
Our aim is to prove Theorem 3.10 which gives an explicit formula for \(G_\lambda \) (3.14) in terms of a sum over a pair of permutations. The argument is longer than in the case of \(F_\lambda \) from Appendix A.1 but also involves manipulations with row operators. Namely, we utilize the operators A, B, C, D given by (2.8)–(2.9). They are built from the vertex weights W and depend on x, r and the sequences \({\textbf{y}}, {\textbf{s}}\). These operators act (from the left) on tensor products of two-dimensional spaces \(V^{(k)}=\mathop {\textrm{span}}\{ e_0^{(k)},e_1^{(k)} \} \simeq {\mathbb {C}}^2\), where \(k\ge 1\). Recall (Sect. 3.1) that to \(\lambda \) we associate the vector \(e_{{\mathcal {S}}(\lambda )}\) in the finitary subspace \({\mathscr {V}}\) of the infinite tensor product \(V^{(1)}\otimes V^{(2)}\otimes \ldots \), where we take \(e^{(k)}_1\) in the k-th place if and only if \(k\in {\mathcal {S}}(\lambda )\) and \(e_0^{(k)}\) otherwise, see Sect. 3.1. Let us also set
Equip all tensor products of the spaces \(V^{(k)}\) with the inner product defined by \(\langle e_{{\mathcal {T}}},e_{{\mathcal {T}}'} \rangle = {\textbf{1}}_{{\mathcal {T}}={\mathcal {T}}'}\) (here we use the notation \(e_{{\mathcal {T}}}\) as in (3.2)). Then by Proposition 3.4 we have
We will compute the above coefficient of \(e_{{\mathcal {S}}(\lambda )}\) in the action of the product of the D operators using the Yang–Baxter equation stated in Proposition 2.4 as a series of commutation relations between the operators A, B, C, and D.
Remark A.5
Sometimes, to shorten some formulas in the proofs, we will use notation \(A_i,B_i,C_i\), or \(D_i\) for \(A(x_i,r_i),B(x_i,r_i),C(x_i,r_i)\), and \(D(x_i,r_i)\), respectively.
1.2.2 Action of D operators on a two-fold tensor product
The next two statements, Lemmas A.6 and A.7, are parallel to the computations with the row operators performed in Appendix A.1.2 in the proof of the formula for \(F_\lambda \).
Lemma A.6
Let \(\sigma \in {\mathfrak {S}}_M\) be a permutation. Then
Proof
This is proven by induction on the length of the permutation \(\sigma \) using the commutation relation (2.11) between the C operators. \(\square \)
Lemma A.7
As operators on a tensor product of two spaces \(V_1\otimes V_2\), we have
Here \({\mathcal {I}}=(i_1<\ldots <i_k )\) and \({\mathcal {J}}=\left\{ 1,\ldots ,M \right\} {\setminus } {\mathcal {I}}= (j_1<\ldots <j_{M-k} )\).
Proof
In the proof we use the shorthand notation for the operators from Remark A.5. By the last identity in (2.9), the action of \(D_M\ldots D_1\) on \(V_1\otimes V_2\) is given by
where \(X_{{\mathcal {K}}}=X_M({\mathcal {K}})\ldots X_1({\mathcal {K}}) \), \(Y_{{\mathcal {K}}}=Y_M({\mathcal {K}})\ldots Y_1({\mathcal {K}}) \), with
Next, by repeated use of relations (2.15) and (2.17), the sum (A.9) can be expressed in the form
where \(h_{I;I'}({\textbf{x}};{\textbf{r}})\) are rational functions in \({\textbf{x}}=(x_1,\ldots ,x_M )\) and \({\textbf{r}}=(r_1,\ldots ,r_M )\), and the indices are
By looking at relations (2.15), (2.17) closer, one can already see that \(m=M-k\) in (A.10). By Remark A.1, the coefficients \(h_{I;I'}({\textbf{x}};{\textbf{r}})\) are independent of the order in which we apply the commutation relations between the operators A, B, C, D to get from (A.9) to (A.10).
By the same argument as in Lemma A.2, one can show that \(h_{I;I'}({\textbf{x}};{\textbf{r}})=0\) if \(I\cap I'\ne \varnothing \) or \(J\cap J'\ne \varnothing \). Thus, it must be that \(I=J'\) and \(J=I'\), and we may rewrite \(h_I({\textbf{x}};{\textbf{r}})=h_{I;I'}({\textbf{x}};{\textbf{r}})\). This implies that we may write (A.10) as
It remains to evaluate the coefficients \(h_I({\textbf{x}};{\textbf{r}})\) in (A.11). This is simpler than for the case of \(F_\lambda \) considered in Appendix A.1.2. First, assume that \(I=I_k:=\left\{ 1,2,\ldots ,k \right\} \). In this case, applying (2.15) and (2.17) to a term \(X_{{\mathcal {K}}} \otimes Y_{{\mathcal {K}}}\) in (A.9) only gives rise to a nonzero multiple of \(B_{k} \cdots B_{1} D_{M} \cdots D_{k+1} \otimes D_{M} \cdots D_{k+1} C_{k} \cdots C_{1}\) as a summand only if \({\mathcal {K}} = I_k\). Indeed, otherwise let \(k_0=\min {\mathcal {K}}^c\le k\). In any expression of \(X_{{\mathcal {K}}} \otimes Y_{{\mathcal {K}}}\) as a linear combination of \(B_{i_k} \cdots B_{i_1} D_{j_{M - k}} \cdots D_{j_1} \otimes D_{j_{M - k}} \cdots D_{j_1} C_{i_k} \cdots C_{i_1}\), one needs to commute \(D_{k_0}\) to the right through \(X_{k_0-1},\ldots ,X_1 \), which implies that \(j_1\le k_0\le k\). Therefore, it must be \({\mathcal {K}}=I_k\).
For \({\mathcal {K}}=I_k\), the only way of obtaining \(B_{k} \cdots B_{1} D_{M} \cdots D_{k+1} \otimes D_{M} \cdots D_{k+1} C_{k} \cdots C_{1}\) from \(X_{{\mathcal {K}}} \otimes Y_{{\mathcal {K}}}\) is through using (2.15) to commute each \(D_j\) to the right of each \(B_i\). This produces a factor of \((r_i^{-2}x_i-x_j)/(r_i^{-2}x_i-r_j^{-2}x_j)\) for each such commutation, and so
Finally, to get \(h_I\) for general I, observe that the operators \(D_i\) commute by (2.13), and therefore \(h_{\sigma (I_k)}({\textbf{x}};{\textbf{r}})=h_{I_k}(\sigma ({\textbf{x}});\sigma ({\textbf{r}}))\), which are precisely the coefficients in the claimed identity in the present lemma, where \(\sigma \) takes \(I_k\) to an arbitrary I. This completes the proof. \(\square \)
For the next proposition, recall the notation \(d = d(\lambda ) \ge 0\) which is the integer such that \(\lambda _d \ge d\) and \(\lambda _{d + 1} < d + 1\), and \(\mu = (\mu _1< \mu _2< \ldots < \mu _d) = \{1,\ldots ,N \} {\setminus } \big ( {\mathcal {S}}(\lambda ) \cap \{1,\ldots ,N \} \big )\). Also consider the N-fold tensor product \(V^{(1)}\otimes \ldots \otimes V^{(N)}\), and take the following vectors in this space
where \(m_i={\textbf{1}}_{i\in {\mathcal {S}}(\lambda )}\), and with \(e_{[1,N]}\) we are slightly abusing the notation, cf. (A.7).
Proposition A.8
With the above notation, for any vectors \(v_1,v_2\in V^{(N+1)}\otimes V^{(N+2)}\otimes \ldots \) we have
where \({\mathcal {I}}=(i_1<\ldots <i_d )\).
The right-hand side of (A.13) vanishes if \(d(\lambda )>M\). Observe that the same is true for the left-hand side. Indeed, a single D operator moves at most one vertical arrow somewhere to the right, and d is the number of gaps (sites with no vertical arrows) among \(\left\{ 1,\ldots ,N \right\} \) in the configuration encoded by \(e_{{\mathcal {S}}_N(\lambda )}\), so d should not be larger than M.
Proof of Proposition A.8
In this proof we use the shorthand notation for the operators, see Remark A.5. As a first step, we consider how the action of the product of the D and C operators like in the right-hand side of (A.13) acts on tensor products. Fix an integer \(n>0\), a subset \({\mathcal {H}}=\left\{ h_1,h_2,\ldots ,h_k \right\} \subseteq \left\{ 1,\ldots ,M \right\} \), and \(u,w\in V^{(n+1)}\otimes V^{(n+2)}\otimes \ldots \). Then we have
and
Indeed, observe that C(x, r) maps \(e_1^{(n)}\) to 0, so by the third statement in (2.9) we have
When applying a product of the \(D_j\)’s to this vector, a nonzero term with \(e_1^{(n)}\) in the first tensor factor may appear only if we act each time by the operators D on both tensor factors, see the fourth statement in (2.9). This (together with the fact that \(\langle \cdot ,\cdot \rangle \) is multiplicative with respect to the tensor product) leads to (A.14). For (A.15), we use Lemma A.7 expressing the action of a product of the \(D_j\)’s on a tensor product, and observe that a nonzero term with \(e_0^{(n)}\) in the first tensor factor may appear only if \(|{\mathcal {I}}| =1\) in the right-hand side of (A.8).
The action of all the operators on \(e_1^{(n)}\) in the right-hand sides of (A.14)–(A.15) is explicit by (2.8) and (2.3):
This means that we can continue our identities as
Now we can evaluate
by repeatedly using (A.16). Start with \({\mathcal {H}}=\varnothing \), and apply the first identity in (A.16) for each \(n\notin \mu =\left\{ 1,\ldots ,N \right\} {\setminus } {\mathcal {S}}(\lambda )\), and the second identity in (A.16) for each \(n\in \mu \). Each application of the latter involves choosing an index \(i\notin {\mathcal {H}}\). This freedom is encoded by the data \(({\mathcal {I}},\sigma )\), where \({\mathcal {I}}=\{i_1<i_2< \ldots < i_d \}\subseteq \left\{ 1,\ldots ,M \right\} \) and \(\sigma \in {\mathfrak {S}}_d\), such that at each step when \(n=\mu _k\in \mu \) we remove the index \(i_{\sigma (k)}\). For each fixed \(({\mathcal {I}},\sigma )\) we have the following factors in the resulting expansion:
-
The inner product term \(\displaystyle \Bigl \langle v_2, \Bigl ( \prod \limits _{j\notin {\mathcal {I}}} D_j\Bigr ) C_{i_d}\ldots C_{i_1} v_1\Bigr \rangle \prod \limits _{1\le \upalpha<\upbeta \le d:\sigma (\upbeta )<\sigma (\upalpha )} \frac{r_{i_{\sigma (\upbeta )}}^{-2} x_{i_{\sigma (\upbeta )}}- x_{i_{\sigma (\upalpha )}}}{r_{i_{\sigma (\upalpha )}}^{-2} x_{i_{\sigma (\upalpha )}}-x_{i_{\sigma (\upbeta )}}} \), where the last factor comes from reordering the C operators thanks to Lemma A.6.
-
The factor \(\displaystyle \prod \limits _{i\in {\mathcal {I}},\, j\notin {\mathcal {I}}} \frac{r^{-2}_i x_i-x_j}{r^{-2}_i x_i - r^{-2}_j x_j} \prod \limits _{1\le \upalpha <\upbeta \le d} \frac{r^{-2}_{i_{\sigma (\upalpha )}} x_{i_{\sigma (\upalpha )}}-x_{i_{\sigma (\upbeta )}}}{r^{-2}_{i_{\sigma (\upalpha )}} x_{i_{\sigma (\upalpha )}} - r^{-2}_{i_{\sigma (\upbeta )}} x_{i_{\sigma (\upbeta )}}}\) arises by applying the second identity in (A.16) for each \(n\in \mu \). Reordering the denominator in the second factor gives
$$\begin{aligned} \prod \nolimits _{1\le \upalpha<\upbeta \le d} \frac{1}{r^{-2}_{i_{\sigma (\upalpha )}} x_{i_{\sigma (\upalpha )}} - r^{-2}_{i_{\sigma (\upbeta )}} x_{i_{\sigma (\upbeta )}}} =\mathop {\textrm{sgn}}(\sigma ) \prod \nolimits _{i,j\in {\mathcal {I}},\, i<j} \frac{1}{r_i^{-2}x_i-r_j^{-2}x_j}. \end{aligned}$$ -
The product \(\displaystyle \prod \nolimits _{j=1}^d \frac{s_{\mu _j}^2 x_{i_{\sigma (j)}}(r_{i_{\sigma (j)}}^{-2}-1)}{y_{\mu _j}-s_{\mu _j}^2x_{i_{\sigma (j)}}}\) is composed of one factor per each application of the second identity in (A.16) corresponding to \(n=\mu _j\in \mu \).
-
The product \(\displaystyle \prod \nolimits _{j=1}^{d}\prod \nolimits _{n=\mu _j+1}^N \frac{s_n^2(x_{i_{\sigma (j)}}-r_{i_{\sigma (j)}}^2 y_n)}{r_{i_{\sigma (j)}}^2(y_n-s_n^2x_{i_{\sigma (j)}})} \) arises from both identities in (A.16) which contain the same products over \(k\in {\mathcal {H}}\).
-
Finally, the product \(\displaystyle \biggl ( \prod \nolimits _{n=1}^N\prod \nolimits _{j=1}^{M} \frac{y_n-s_n^2r_j^{-2} x_j}{y_n-s_n^2x_j} \biggr ) \biggl ( \prod \nolimits _{j=1}^d \prod \nolimits _{n=\mu _j}^{N} \frac{y_n-s_n^2x_{i_{\sigma (j)}}}{y_n-s_n^2r_{i_{\sigma (j)}}^{-2} x_{i_{\sigma (j)}}} \biggr ) \) arises from the products over \(j\notin {\mathcal {H}}\) or \(j\notin {\mathcal {H}}\cup \left\{ i \right\} \) in (A.16).
Combining all the terms yields the desired identity.
1.2.3 Commutation of the operators C and D
In this subsection we establish one of the key formulas concerning the commutation of the operators C and D. We fix \(M,N\ge 1\) and sequences of complex numbers
Proposition A.9
We have
Here \({\mathcal {I}}=(i_1<\ldots <i_k )\) and \({\mathcal {H}}=(h_1<\ldots <h_{M-k} )\).
Recall that the operators \(D(x_j,r_j)\) commute by (2.13), so we can write their products in any order. This is not the case for the operators \(C(w_j,\theta _j)\), which is why their order in (A.17) must be specified explicitly.
The rest of this subsection is devoted to the proof of Proposition A.9. As a first step, let us establish the claim for \(M=1\):
Lemma A.10
(Proposition A.9 for \(M=1\)) We have
Proof
The first term containing \(C(w,\theta ) D(x_1,r_1)\ldots D(x_N,r_N)\) may only arise if we are picking the first summand in (2.16) for each commutation. This produces the desired product \(\prod _{j=1}^{N}\frac{r_j^{-2}x_j-w}{x_j-w}\) as a prefactor.
Now let us explain how to get the summand in the second sum corresponding to \(i=1\). Thanks to the commutativity of the \(D(x_j,r_j)\)’s, the form of the other summands then would follow. To get the term containing \(C(x_1,r_1)D(w,\theta )D(x_2,r_2)\ldots D(x_N,r_N)\), we must pick the second summand in (2.16) once, when moving \(C(w,\theta )\) to the left of \(D(x_1,r_1)\). This produces \(C(x_1,r_1)D(w,\theta )\frac{(1-r_1^{-2})x_1}{x_1-w}\). After that, we move \(C(x_1,r_1)\) to the left of all the other \(D(x_j,r_j)\)’s, always picking the first summand in (2.16). This produces the desired identity.
We now consider the general case \(M,N\ge 1\) of (A.17). First, repeatedly using relations (2.11), (2.13), and (2.16), we have
where the sum is taken over \({\mathcal {I}}\subseteq \left\{ 1,\ldots ,N \right\} \) and \({\mathcal {H}}\subseteq \left\{ 1,\ldots ,M \right\} \), such that \(|{\mathcal {I}}| =k\), \(|{\mathcal {H}}| =M-k\), and k is arbitrary (see (A.17)). Here \(R_{{\mathcal {I}};{\mathcal {H}}}\) are some rational functions which we will now evaluate.
Lemma A.11
(Evaluation of \(R_{{\mathcal {I}};{\mathcal {H}}}\) in a special case) Let \({\mathcal {H}}=\{1,2,\ldots ,M-k \}\), and \({\mathcal {I}}=(i_1<\ldots <i_k )\subseteq \left\{ 1,\ldots ,N \right\} \) with \(|{\mathcal {I}}| =k\) be arbitrary. Then
Proof
From the left-hand side of (A.19), we apply (2.16) (together with permutation relations (2.11), (2.13) for the operators C, D) to move all the operators C to the left of all the operators D. The operator
may arise, after a sequence of applications of Lemma A.10, only if there exists a permutation \(\sigma \in {\mathfrak {S}}_k\) such that the following two conditions are met:
-
When moving each \(C(w_{M-k+j},\theta _{M-k+j})\), \(1\le j\le k\), to the left, turn \((w_{M-k+j},\theta _{M-k+j})\) into \((x_{i_{\sigma (j)}},r_{i_{\sigma (j)}})\). This corresponds to picking the second summand in (2.16), and this swapping of parameters may happen only once per each C operator.
-
When moving each \(C(w_j,\theta _j)\), \(1\le j\le M-k\), to the left, we always pick the first summand in (2.16), and the parameters \((w_j,\theta _j)\) stay the same throughout the exchanges.
To be able to put all the coefficients together, denote \(\sigma _t ({\mathcal {I}}) = \big ( i_{\sigma (t)}, i_{\sigma (t + 1)}, \ldots , i_{\sigma (k)} \big )\) for each \(1\le t\le k\). Then, for each integer \(1\le j\le k\), when attempting to commute \(C(w_{M - k + j}, \theta _{M - k + j})\) to the left of
we obtain
By Lemma A.10, this contributes a factor of
This deals with the first case above when we swap the parameters between C and D operators.
In the second case when we do not swap the parameters, each \(C (w_j, \theta _j)\) for \(1\le j\le M-k\) must be commuted to the left of \(\prod _{h \notin {\mathcal {I}}} D (x_h, r_h) \prod _{h = M - k + 1}^M D (w_h, \theta _h)\), which contributes
Observe that
Now, combining the product of (A.21) over \(1\le j\le k\) and (A.22) over \(1\le j\le M-k\), and using (A.23), we see that the desired coefficient depending on \(\sigma \in {\mathfrak {S}}_k\) is equal to
Note that this is the coefficient of the operator
and permuting the first k of the C operators to the desired order \(C(x_{i_k},r_{i_k})\ldots C(x_{i_1},r_{i_1}) \) results in an additional factor
by Lemma A.6.
This implies that the full coefficient \(R_{{\mathcal {I}};{\mathcal {H}}}({\textbf{w}};{\textbf{x}};\varvec{\uptheta };{\textbf{r}})\) equals to the sum of (A.24) times (A.25) over all \(\sigma \in {\mathfrak {S}}_k\). We have
Therefore, the summation over \(\sigma \) amounts to computing the determinant:
We have already computed this determinant (up to renaming the variables) in (3.9), and so
where we recalled that \({\mathcal {H}}=\left\{ 1,2,\ldots ,M-k \right\} \). Combining this with the remainder of (A.24), we arrive at the desired expression (A.20), thus concluding the proof of Lemma A.11. \(\square \)
Finally, to get \(R_{{\mathcal {I}};{\mathcal {H}}}\) for general \({\mathcal {H}}\), we can permute the C operators in the left-hand side of (A.17) thanks to (2.11). More precisely, the two expressions
are symmetric in \((w_i,\theta _i)\), \(1\le i\le M\), and \((w_h,\theta _h)\), \(h\in {\mathcal {H}}\), respectively. Defining
we see that for any permutation \(\tau \in {\mathfrak {S}}_M\) we have \( {\widehat{R}}_{{\mathcal {I}};\tau ({\mathcal {H}})}(\tau ({\textbf{w}});{\textbf{x}};\tau (\varvec{\uptheta });{\textbf{r}}) = {\widehat{R}}_{{\mathcal {I}};{\mathcal {H}}}({\textbf{w}};{\textbf{x}};\varvec{\uptheta };{\textbf{r}}) \). The renormalization in (A.27) cancels out with the two last factors in \(R_{{\mathcal {I}};\left\{ 1,\ldots ,M-k \right\} }\) in (A.20). This together with the symmetry of (A.27) implies that \(R_{{\mathcal {I}};{\mathcal {H}}}\) for general \({\mathcal {H}}\) is given by the same formula. We have thus completed the proof of Proposition A.9.
1.2.4 Action of C operators on a two-fold tensor product
In this subsection we perform computations with row operators acting on tensor products which are parallel to those in Appendices A.1.2 and A.2.2, but now involve the C operators.
Lemma A.12
Let \({\textbf{x}}=(x_1,\ldots ,x_M )\), \({\textbf{r}}=(r_1,\ldots ,r_M )\). On any tensor product \(V_1\otimes V_2\) we have:
where \({\mathcal {I}}=(i_1<\ldots <i_k )\) and \({\mathcal {J}}=\left\{ 1,\ldots ,M \right\} {\setminus } {\mathcal {I}}= (j_1<\ldots <j_{M-k} )\).
Proof
In the proof we use the shorthand notation for the operators from Remark A.5. Due to (2.9), relations in Proposition 2.4, and an argument identical to the beginning of the proof of Lemma A.7, we see that the left-hand side of (A.28) can be written in the form
where the notation \({\mathcal {I}},{\mathcal {J}}\) is as in (A.28).
We first evaluate \(h_{{\mathcal {I}}}\) in the special case \({\mathcal {I}}={\mathcal {I}}_k=\left\{ M-k+1,\ldots ,M-1,M \right\} \). The contribution containing the operator \(C_M\ldots C_{M-k+1}A_{M-k}\ldots A_1\otimes C_{M-k}\ldots C_1 D_{M}\ldots D_{M-k+1}\) may arise only if we use (2.16) in the second tensor factor to commute all \(C_j\), \(j\notin {\mathcal {I}}_k\), to the left of all \(D_i\), \(i\in {\mathcal {I}}\), without swapping their arguments. Each such commutation gives rise to the factor \(\frac{r_i^{-2}x_i-x_j}{x_i-x_j}\). Therefore,
Next, thanks to (2.11) the three expressions
are symmetric in the pairs \((x_i,r_i)\) of variables they depend on (where \(1\le i\le M\), \(i\in {\mathcal {I}}\), and \(i\notin {\mathcal {I}}\), respectively). Therefore, the function
satisfies \({{\widehat{h}}}_{\tau ({\mathcal {I}})}({\textbf{x}};{\textbf{r}}) ={{\widehat{h}}}_{{\mathcal {I}}}(\tau ^{-1}({\textbf{x}});\tau ^{-1}({\textbf{r}}))\) for any permutation \(\tau \in {\mathfrak {S}}_M\). Together with (A.29) this shows that for any \({\mathcal {I}}\) we have \({{\widehat{h}}}_{{\mathcal {I}}}({\textbf{x}};{\textbf{r}})=\prod _{i\in {\mathcal {I}},\,j\notin {\mathcal {I}}}(x_i-x_j)^{-1}\), which implies the claim. \(\square \)
In the next proposition, let \(e_0 = e_0^{(i_1)} \otimes e_0^{(i_2)} \otimes \cdots \otimes e_0^{(i_n)} \in V^{(i_1)} \otimes V^{(i_2)} \otimes \cdots \otimes V^{(i_n)}\) for any integers \(i_1< i_2< \cdots < i_n\). Moreover, fix \(M \ge 1, N \ge 0\), and \({\mathcal {T}} = (t_1< t_2< \ldots < t_M) \subset {\mathbb {Z}}_{\ge 1}\). Define the vector \(e_{{\mathcal {T}}; N} = e_{m_1}^{(N + 1)} \otimes e_{m_2}^{(N + 2)} \otimes \cdots \in V^{(N + 1)} \otimes V^{(N + 2)} \otimes \cdots \), where \(m_i=1\) if \(i\in {\mathcal {T}}\), and 0 otherwise.
Proposition A.13
With the above notation we have
where the inner product is taken in the space \(V^{(N+1)}\otimes V^{(N+2)}\otimes \ldots \).
Observe that this formula is determinantal, and is in fact equivalent to the determinantal formula for \(F_\lambda \) from Theorem 3.9 proven in Appendix A.1, up to swapping horizontal arrows with empty horizontal edges, and renormalizing. Here, however, we present an independent proof which is more convenient given our previous statements.
Proof of Proposition A.13
In the proof we use the shorthand notation for the operators from Remark A.5. Fix \(n>N\) and vectors \(v_1,v_2\in V^{(n+1)}\otimes V^{(n+2)}\otimes \ldots \). By Lemma A.12, we have
These quantities can be computed as follows:
using the definition of the operators (2.8) and formulas for the vertex weights W (2.3). Therefore, (A.30) is continued as
Now we can evaluate \(\langle e_{{\mathcal {T}};N},C_M \ldots C_1 e_0 \rangle \) by repeatedly applying (A.31). Throughout these applications, we use first or second identity in (A.31), respectively, for each n belonging or not belonging to the set \(\left\{ t_1+N,t_2+N,\ldots ,t_M+N \right\} \). In the latter case, for \(n=N+t_j\), we choose which index \(i=i_j\in \left\{ 1,\ldots ,M \right\} \) to remove. These choices are encoded by a permutation \(\sigma \in {\mathfrak {S}}_M\) as \(i_j=\sigma (j)\). This leads to the desired claim, where, in particular, \(\mathop {\textrm{sgn}}(\sigma )\) arises from reordering the denominators \(x_{\sigma (i)}-x_{\sigma (j)}\) to \(x_i-x_j\) over all \(1\le i<j\le M\). \(\square \)
1.2.5 Completing the proof
To finalize the proof of Theorem 3.10, let us recall the formula to be established. Fix an arbitrary signature \(\lambda =(\lambda _1\ge \ldots \ge \lambda _N \ge 0)\). Let \(d = d(\lambda ) \ge 0\) denote the integer such that \(\lambda _d \ge d\) and \(\lambda _{d + 1} < d + 1\). Denote by \(\ell _j=\lambda _j+N-j+1\), \(j=1,\ldots ,N \), the elements of the set \({\mathcal {S}}(\lambda )\). Moreover, we define \(\mu = (\mu _1< \mu _2< \ldots < \mu _d) = \{1,\ldots ,N \} {\setminus } \big ( {\mathcal {S}}(\lambda ) \cap \{1,\ldots ,N \} \big )\). Our goal is to show that
where \({\mathcal {I}}= (i_1< i_2< \ldots < i_d)\) and \({\mathcal {J}}= (j_1< j_2< \ldots < j_d)\).
Recall that
where we have split the vectors into \(e_{{\mathcal {S}}_N(\lambda )},e_{[1,N]}\in V^{(1)}\otimes \ldots \otimes V^{(N)}\) (cf. (A.12)), and the remaining two vectors belong to \(V^{(N+1)}\otimes V^{(N+2)}\otimes \ldots \). Note that the vector \(e_{{\mathcal {S}}_{>N}(\lambda )}\) has exactly d tensor components of the form \(e_1^{(k)}\), and the other components are of the form \(e_0^{(k)}\). We can use Proposition A.8 to write:
Let us denote
and use similar notation in what follows. In particular, in all such products of the C operators the indices are decreasing from left to right. Employ Proposition A.9 to write
Let us insert this into (A.33). Observe that all operators D preserve the vector \(e_0\). Thus, we can continue the computation as
Now we are going to apply Proposition A.13 to compute the remaining inner product. Recall that \(e_{{\mathcal {S}}_{>N}(\lambda )}\) has exactly d tensor components equal to \(e_1^{(m)}\), for \(m \in \left\{ \ell _1,\ldots ,\ell _d \right\} \). Denote \((x_1',\ldots ,x_d' )=(x_{h_1},\ldots ,x_{h_{|{\mathcal {H}}|}}, x_{k_1},\ldots ,x_{k_{|{\mathcal {K}}|}} )\), where \(h_1<\ldots <h_{|{\mathcal {H}}|} \), \(k_1<\ldots <k_{|{\mathcal {K}}|} \). Then we have
The sign \((-1)^{\frac{d(d-1)}{2}}\) arises from the fact that the \(t_j\)’s in Proposition A.13 are increasing, while the \(\ell _j\)’s in (A.34) are decreasing, so the sign of \(\rho \) has to be multiplied by \((-1)^{\frac{d(d-1)}{2}}\). This allows to continue our computation as follows:
Upon denoting \({\mathcal {J}}={\mathcal {K}}\cup {\mathcal {H}}=(j_1<\ldots <j_d )\), we arrive at the desired statement (A.32). Note that reordering the indices in \((x_1',\ldots ,x_d' )\) in the increasing order leads to an extra ± sign coming from \(\mathop {\textrm{sgn}}(\rho )\), but this sign is compensated by writing
(equivalently, one may refer to the symmetry as in the proof of Lemma A.12). Finally, replacing \(x_i-x_j\) in (A.35) with \(x_j-x_i\) absorbs the sign \((-1)^{\frac{d(d-1)}{2}}\). This completes the proof of Theorem 3.10.
Correlation kernel via Eynard–Mehta approach
Here we prove Theorem 6.7 on the determinantal structure of the FG measures and processes. We employ an Eynard–Mehta type approach based on [20], see also [31].
1.1 Representation of the ascending FG process in a determinantal form
Recall the notation of the ascending FG process (6.8) from Sect. 6.2. Throughout Appendix B we omit the notation \({\textbf{y}},{\textbf{s}}\) in the functions \(G_{\mu /\varkappa }(w_i;{\textbf{y}};\theta _i;{\textbf{s}})\) and other similar quantities.
Here we use the determinantal formulas for the functions \(F_\lambda \) (Theorem 3.9) and \(G_\mu \), \(G_{\nu /\lambda }\) to rewrite the probabilities (6.8) in a determinantal form. The formulas for \(G_\mu \) and \(G_{\nu /\lambda }\) are of Jacobi–Trudy type and follow from Cauchy identities and biorthogonality as in Sect. 5.4.
Recall the notation (3.11):
By Theorem 3.9, we have
where the constant is independent of \(\lambda \) (we adopt this convention for all such constants throughout Appendix B, and will denote all of them by \(\textrm{const}\)).
Next, recall the functions \(\psi _k\) (5.1):
For \(({\textbf{w}};\varvec{\uptheta })=(w_a,\ldots ,w_b;\theta _a,\ldots ,\theta _b)\), \(a\le b\), let us define a slight generalization of (5.13):
where the integration contour \(\Gamma _{y,w}\) is positively oriented, surrounds all \(y_i, w_j\), and leaves out all \(s_i^{-2}y_i\). The function \(G_{\lambda ^{(1)}}\) in (6.8) has the following determinantal form (with \(a=b=1\) in \({\textsf{h}}_{k,l}\)):
Finally, recall the functions \(\widetilde{{\textsf{h}}}_l\) (5.9) and \({\textsf{g}}_{l/k}\) (5.10):
where the integration contour is around \(y_j,w_i\) and not \(s_j^{-2}y_j\), and \(({\textbf{w}};\varvec{\uptheta })=(w_a,\ldots ,w_b;\theta _a,\ldots ,\theta _b)\) with \(a\le b\). The skew functions in (6.8) take the following determinantal form:
We observe that when evaluated at a single pair of variables \((w;\theta )\), both \({\textsf{h}}_{k,j}\) (for \(k\ge j\)) and \({\textsf{g}}_{l/k}\) become explicit:
Lemma B.1
We have
Proof
For \({\textsf{h}}_{k,j}\) with \(k\ge j\), the only pole inside the contour is \(z=w\), which leads to the desired formula. The exact form of the functions \({\textsf{h}}_{k,j}\) with \(k < j\) is not very explicit (apart from the original contour integral expression), but they are not involved in our computations.
For \({\textsf{g}}_{l/k}\), in the case \(l=k\), the only singularity outside the contour is \(z=s_{k+1}^{-2}y_k\), and for \(l>k\) the only singularity inside the contour is \(z=w\). The respective residues in these two cases lead to the desired formulas. For \(l<k\), there are no singularities outside the integration contours, and the integral vanishes. \(\square \)
Putting together (B.1), (B.2), and (B.3), we get:
Proposition B.2
The probability weights under the ascending FG process (6.8) have the following product-of-determinants form. For \(\ell ^{(t)}_j:=\lambda ^{(t)}_{j}+N+1-j\), we have
where all determinants are taken with respect to \(1\le i,j\le N\), and \(\textrm{const}\) is a normalizing constant which does not depend on the \(\ell ^{(j)}\)’s.
1.2 Application of the Eynard–Mehta theorem
The form of the probability weights as in Proposition B.2 puts the ascending FG process into the domain of applicability of the Eynard–Mehta theorem (see, for example, [31, 20, Theorem 1.4]). To express the determinantal correlation kernel of the point process
one first needs to invert the \(N\times N\) “Gram matrix” given by
Note that by Lemma B.1, this series converges absolutely under the condition (6.7).
Proposition B.3
We have
The proof is based on the following lemma:
Lemma B.4
Let \(\bigl | \frac{u-s_j^{-2}y_j}{u-y_j} \frac{v-y_j}{v-s_j^{-2}y_j} \bigr |<1-\delta <1\) for all sufficiently large \(j\ge 1\). Then we have
Proof
We have
and the sum telescopes to \(1 / (u-v)\) if it converges (which holds under the condition in the hypothesis). \(\square \)
Proof of Proposition B.3
We represent \({\textsf{h}}_{a_1,i}\) as an integral over \(z_1\), and each \({\textsf{g}}_{a_t/a_{t-1}}\) as an integral over \(z_t\), \(2\le t\le T\). Initially all the integration variables belong to the same contour \(\Gamma _{y,w}\). However, in order to apply Lemma B.4 under the integrals, we need to have the following conditions on the contours for all sufficiently large \(k\ge 1\):
where \(t=1,\ldots ,T-1\), \(j=1,\ldots ,N \). Clearly, under certain restrictions on the parameters, such contours exist. Moreover, we may also choose them to be nested: \(z_1\) around all \(y_k\) and \(w_t\), \(z_{B}\) around \(z_A\) if \(B>A\), and all contours must leave outside all the points \(s_k^{-2}y_k\). On these contours, we have by Lemma B.4:
This integral is computed as follows. First, for \(z_T\) there is a single pole \(z_T=x_j\) outside the contour (and the integrand has the zero residue at infinity). Taking the residue clears the denominator \(x_j-z_T\) and substitutes \(z_T=x_j\). After that, we repeat the procedure for \(z_{T-1},\ldots ,z_1\), which leads to the desired formula.
Finally, the restrictions on the parameters under which the contours exist are lifted by an analytic continuation, since Lemmas B.1 and B.4 imply that the summation in (B.5) produces an a priori rational function. \(\square \)
The matrix \(M=[M_{ij}]_{i,j=1}^{N}\) is readily inverted:
Lemma B.5
We have, for \(i,j=1,\ldots ,N\),
where the contours for \(\xi \) and \(\eta \) are small nonintersecting positively oriented circles around \(x_i\) and \(y_j\), respectively, which do not include any other poles of the integrand.
Proof
The first expression for \(M_{ij}^{-1}\) is obtained using the Cauchy determinant, since all minors (and hence all cofactors) of M are determinants of similar form. The contour integral expression corresponds to taking residues at the simple poles \(\xi =x_i\) and \(\eta =y_j\).
By the Eynard–Mehta theorem as in [20, Theorem 1.4], the correlation kernel of the determinantal point process (B.4) on \(\left\{ 1,\ldots ,T \right\} \times {\mathbb {Z}}_{\ge 1}\) takes the form (the shifts \(a+1,a'+1\) correspond to the shifts in the determinantal representation in Proposition B.2):
The iterated sums over the \(\alpha _j\)’s in the first and the second terms are finite and thus converge, and the sum over the \(\beta _j\)’s is infinite but converges under (6.7), see Lemma B.1.
1.3 Computation of the kernel
Let us now compute all the sums in (B.8), and arrive at the resulting formula for the correlation kernel.
For the first summand arising when \(t>t'\), we pass to the nested contours ( \(z_{B}\) around \(z_A\) if \(B>A\)) as in the proof of Proposition B.3. We obtain
where we extended the sum over the \(\alpha _j\)’s to all \(\alpha _j\ge 0\) under the integral, and the infinite sums under the integral are computed using Lemma B.4. Next, deforming the contours \(z_{t-1},z_{t-2},\ldots ,z_{t'+1} \) (in this order) to infinity, each integration in \(z_i\) picks up a residue at a single pole outside the integration contour at \(z_i=z_t\). This leaves a single integral:
Arguing in a similar manner, we can compute
and
In the latter computation we pick the residues at \(z_T=x_j,\ldots ,z_{t'+1}=x_j\) (in this order), which is the only pole outside the corresponding integration contour. Finally, we take the last two quantities, multiply by \(M_{ji}^{-1}\), and sum as in (B.8). Using (B.7), we have
To obtain the latter expression we substituted \(x_j=\xi \), \(y_i=\eta \), and changed the contours \(\Gamma _x,\Gamma _y\) for these variables to encircle all \(x_k\)’s or all \(y_k\)’s, respectively, while leaving all other poles outside. Observe now that the only pole in \(\eta \) outside the integration contour which produces a nonzero residue is at \(\eta =z\). Indeed, the residue at \(\eta =\xi \) eliminates all poles inside the \(\xi \) contour, and thus vanishes. Therefore, we may continue the above computation as follows:
Let us drag the \(\xi \) contour through infinity, so that now it encircles the z contour \(\Gamma _{y,w}\), and also all the points \(\theta _i^{-2}w_i\). This leads to an extra minus sign.
Finally, we need to add the additional summand (B.10) if \(t>t'\). In this case, observe that dragging the z contour so that it is outside of the \(\xi \) contour produces the same expression as (B.10), but with the opposite sign. Moreover, we need to undo the shifts \(a+1,a'+1\) corresponding to the determinantal representation in Proposition B.2. Renaming the integration variables as \(\xi =u\), \(z=v\) leads to the final expression for the correlation kernel of the ascending FG process:
where the u contour is outside for \(t\le t'\), and the v contour is outside for \(t>t'\). This completes the proof of Theorem 6.7 in the ascending FG process case.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Aggarwal, A., Borodin, A., Petrov, L. et al. Free fermion six vertex model: symmetric functions and random domino tilings. Sel. Math. New Ser. 29, 36 (2023). https://doi.org/10.1007/s00029-023-00837-y
Accepted:
Published:
DOI: https://doi.org/10.1007/s00029-023-00837-y