Arxiv:2106.13378V1 [Math.CO] 25 Jun 2021 the .Cmiaoiso Vlaodn and Pol Evil-Avoiding Schubert of and Combinatorics Permutations Partitions, on 3

arXiv:2106.13378v1 [math.CO] 25 Jun 2021 the .Cmiaoiso vlaodn and pol evil-avoiding Schubert of and Combinatorics permutations partitions, on 3. Background 2. Introduction 1. .Cantini’s 4. .The 5. .Poete of Properties 6. .Mliieqee n taysaepoaiiis27 for formula queue probabilities multiline state A steady and queues 9. Multiline 5.9 Theorem of 8. Proof 7. nrcn er,teehsbe o fwr nitrcigpa interacting on work of lot a been has there years, recent In 5.9 Theorem of proof the for results technical conjecture References Appendix: factor monomial the questions 12. 1.3, Future Theorem of proof 11. The 10. CUETPLNMAS H NOOEEU TASEP, INHOMOGENEOUS THE POLYNOMIALS, SCHUBERT smercsml xlso process exclusion simple asymmetric fweight of obeShbr oyoil.I hspprw nrdc the introduce we paper this permutations In polynomials. Schubert double When usqetygvni em fmliieqee A1,AM13 s [AL14, queues steady y multiline for of terms formulas in given explicit subsequently and [LW12], polynomials bert and w daetprilso weights of particles adjacent two cuidb atce fweights of particles by occupied o the for smercsml xlso process exclusion simple asymmetric htaiei u omlsaeﬂge cu ucin,adgi table and Young Sch semistandard functions, Schur and the queues ﬂagged that multiline are show between formulas case we our Finally, in arise [LW12]. that from conjecture factor nststu orsod othe to corresponds thus sites in Abstract. etrlyalohrpoaiiisaepootoa oapo a When to proportional polynomials.) are Schubert probabilities other all jecturally taysaeprobability state steady in i atn Cn6 hwdthat showed [Can16] Cantini , S n 3214 z n o ahei-viigpermutation evil-avoiding each for and , dfraino u anrsl and result main our of -deformation y z i dfre taysaepoaiiis n s hst prove to this use and probabilities, state steady -deformed eso htteeare there that show We . = j z 0 N VLAODN PERMUTATIONS EVIL-AVOIDING AND st h ih ftepril fweight of particle the of right the to is dfraino taysaepoaiiis13 probabilities state steady of -deformation osdraltieo ie ragdaon ig ihthe with ring, a around arranged sites n of lattice a Consider hc r h emttosaodn h patterns the avoiding permutations the are which , o all for z OGYNKMADLUE .WILLIAMS K. LAUREN AND KIM DONGHYUN Shbr polynomials -Schubert i h ttoaydsrbto a ojcual ikdt S to linked conjecturally was distribution stationary the , ψ w sapouto obeShbr oyoil.(Con- polynomials. Schubert double of product a as 1. y { n n i 1 i ! Introduction fthe of , < Contents = emttosin permutations z 2 ( j n , . . . , dfre rbblts32 probabilites -deformed o AE)i akvcanon chain Markov a is TASEP) (or 0 2 + wppae trate at places swap √ o all for AE) oe nwihprilshpo a on hop particles which in model a (ASEP), n 2 1 k ) ! n Gasana emttos9 permutations -Grassmannian } ttshv rbblte rprinlto proportional probabilities have states − h osbearneet fparticles of arrangements possible the ; 1 +( 2 w i 2 egv utln uu formulas queue multiline give we , egv nepii oml o the for formula explicit an give we , − √ z 2 i Shbr oyoil 15 polynomials -Schubert Ohrientighappens.) nothing (Otherwise . ) S n n − 1 The . vlaodn permutations evil-avoiding iiesmo tlattwo least at of sum sitive x i .I h aeo general of case the In ]. aepoaiiiswere probabilities tate − nooeeu totally inhomogeneous aux. eabjcini this in bijection a ve ls of class y n br polynomials ubert + 1 − tcemdl uhas such models rticle 2413 j nmas7 ynomials h monomial the fteparticle the if evil-avoiding S , n 4132 nwhich in , n , 4213 sites chu- 21 17 48 43 43 38 1 2 DONGHYUNKIMANDLAURENK.WILLIAMS one-dimensional lattice subject to the condition that at most one particle may occupy a given site. The ASEP on a one-dimensional lattice with open boundaries has been linked to Askey-Wilson polynomials and Koornwinder polynomials [USW04, CW11, Can17, CW18], while the ASEP on a ring has been linked to Macdonald polynomials [CdGW15, CMW18]. The inhomogeneous totally asymmetric simple exclusion process (TASEP) is a variant of the exclusion process on the ring in which the hopping rate depends on the weight of the particles. In this paper we build on works of Lam-Williams [LW12], Ayyer-Linusson [AL14], and especially Cantini [Can16] to give formulas for many steady state probabilities of the inhomogeneous TASEP on a ring in terms of Schubert polynomials.

Deﬁnition 1.1. Consider a lattice with n sites arranged in a ring. Let St n denote the n! labelings of the lattice by distinct numbers 1, 2,...,n, where each( number) i is called a particle of weight i. The inhomogeneous TASEP on a ring of size n is a Markov chain with state space St n where at each time t a swap of two adjacent particles may occur: a particle of weight( ) i on the left swaps its position with a particle of weight j on the right with transition rate ri,j given by:

xi yn 1 j if i j ri,j − + − < = 0 otherwise.

In what follows, we will identify each state with a permutation in Sn. Following [LW12, Can16], we multiply all steady state probabilities for St n by the same ( ) constant, obtaining “unnormalized” steady state probabilities ψw, so that

j i 1 (1.1) ψ123...n xi yn 1 j − − Z x; y , = 1 Mi j n( − + − ) ∈ [ ] ≤ < ≤ where x and y denote x1,...,xn 1 and y1,...,yn 1. Figure 1 shows the state diagram and unnormalized steady state probabilities− for −n 3. =

x1 y1 − 123

x2 y1 x1 y2 − − x1 x2 y1 y2 132 213 x1 x2 y1 y2 + − − + − − x1 y1 x1 y1 − − x1 y1 312 231 x1 y1 − x1 y1 − − x1 y2 x2 y1 − − 321 x1 x2 y1 y2 + − − Figure 1. The state diagram for the inhomogeneous TASEP on St 3 , with transition rates shown in blue, and unnormalized steady state( ) probabilities ψw in red. Though not shown, the transition rate 312 → 213 is x2 y1 and the transition rate 231 → 132 is x1 y2. − − SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 3

1 In the case that yi 0, Lam and Williams [LW12] studied this model and con- = jectured that after a suitable normalization, each steady state probability ψw can be written as a monomial factor times a positive sum of Schubert polynomials, see Table 1 and Table 2. They also gave an explicit formula for the monomial factor, and conjectured that under certain conditions on w, ψw is a multiple of a particular Schubert polynomial. Subsequently Ayyer and Linusson [AL14] gave a conjectural combinatorial formula for the stationary distribution in terms of multiline queues, which was proved by Arita and Mallick [AM13]. In [Can16], Cantini introduced the 2 version of the model given in Deﬁnition 1.1 with yi general, and, after introducing a further set of spectral parameters zi, gave a series of exchange equations relating the components ψw z of the z-deformed stationary distribution. This allowed him to give explicit formulas( ) for the steady state probabilities for n of the n! states as products of double Schubert polynomials.

State w Probability ψw 2 1234 x1 y1 x1 y2 x2 y1 ( − ) ( − )( − ) 1324 x1 y1 S1432 ( − ) 1342 x1 y1 x2 y1 S1423 ( − )( − ) 1423 x1 y1 x1 y2 x2 y1 S1243 ( − )( − )( − ) 1243 x1 y2 x1 y1 S1342 ( − )( − ) 1432 S1423S1342 Table 1. The unnormalized steady state probabilities for n 4. =

In this paper we build on [Can16, AL14, AM13] to strengthen the connection between steady state probabilities and Schubert polynomials: in particular, we give a formula for ψw as a “trivial factor” times a product of (double) Schubert polynomials whenever w is evil-avoiding, that is, it avoids the patterns 2413, 4132, 4213 and 3214.3 2 √2 n−1 2 √2 n−1 We show that there are ( + ) +(2 − ) evil-avoiding permutations in Sn, so this gives a substantial generalization of the previous result [Can16] linking probabilities to Schubert polynomials. In the case that yi 0 for all i, we also give a formula for the = z-deformed steady state probabilities ψw z in terms of multiline queues, generalizing the result of Arita and Mallick; we then( use) this result to prove the monomial factor conjecture from [LW12]. Finally, we show that the Schubert polynomials that arise in our formulas are ﬂagged Schur functions, and give a bijection in this case between multiline queues and semistandard Young tableaux.

In order to state our main results, we need a few definitions. First, we say that two states w and w′ are equivalent, and write w w′, if one state is a cyclic shift of the ∼ other, e.g. w1,...,wn w2,...,wn,w1 . Because of the cyclic symmetry inherent in the definition( of the) TASEP∼ ( on a ring,) it is clear that the probabilities of states w and w′ are equal whenever w w′. We will therefore often assume, without loss of ∼ generality, that w1 1. Note that up to cyclic shift, St n contains n 1 ! states. = ( ) ( − ) 1 However the convention of [LW12] was slightly different; it corresponds to labeling states by the inverse of the permutations we use here. 2 We note that in [Can16], the rate ri,j was xi−yj rather than xi−yn 1 j as we use in Definition 1.1. 3 + − We call these permutations evil-avoiding because if one replaces i by 1, e by 2, l by 3, and v by 4, then evil and its anagrams vile, veil and leiv become the four patterns 2413, 4132, 4213 and 3214. (Leiv is a name of Norwegian origin meaning “heir.”) 4 DONGHYUNKIMANDLAURENK.WILLIAMS

State w Probability ψw 6,3,1 12345 x( ) 5,2,0 12354 x( )S13452 4,1,0 12435 x( )S14532 4,1,1 12453 x( )S14523 5,2,1 12534 x( )S12453 3,0,0 12543 x( )S14523S13452 3,1,1 13245 x( )S15423 2,0,0 13254 x( )S15423S13452 3,2,1 13425 x( )S15243 3,3,1 13452 x( )S15234 2,1,0 13524 x( ) S164325 S25431 2,2(,0 + ) 13542 x( )S15234S13452 4,2,0 14235 x( )S13542 4,2,1 14253 x( )S12543 1,0,0 14325 x( ) S1753246 S265314 S2743156 S356214 S364215 S365124 ( + 1+,1,0 + + + ) 14352 x( )S15234S14532 4,3,1 14523 x( )S12534 1,1,1 14532 x( )S15234S14523 5,3,1 15234 x( )S12354 3,1,0 15243 x( ) S146325 S24531 2,1,1 ( + ) 15324 x( ) S15432 S164235 2,2(,1 + ) 15342 x( )S15234S12453 3,2,0 15423 x( )S12534S13452 15432 S15234S14523S13452 Table 2. The unnormalized steady state probabilities for n 5, when a,b,c a b c = each yi 0. In the table, x( ) denotes x x x . = 1 2 3

In Theorem 1.3 below we describe the monomial factors that appear in the steady state probabilities. Suppose that yi 0 for all i. = 1 1 Deﬁnition 1.2. Given w w1,...,wn St n , let r w− i 1 and s w− i . Let = ( ) ∈ ( ) = ( + ) = ( ) αi w be the number of integers greater than i 1 among i 1 wr,wr 1,...,ws i , where( ) the subscripts are taken modulo n. Let η+ w be the{ largest+ = monomial+ that= can} ( ) be factored out of ψw.

Theorem 1.3. [LW12, Conjecture 2] Let yi 0 for all i. For w St n , we have = ∈ ( ) n 2 − αi w αn−2 w η w xi ( )+⋯+ ( ). ( ) = Mi 1 = And if two states w,w′ St n have the same η w η w′ , then w and w′ are ∈ ( ) ( ) = ( ) cyclically equivalent i.e w w′. ∼ See Table 2 for examples. We now introduce some deﬁnitions needed to characterize the Schubert polynomial factors that appear in the probabilities ψw.

Deﬁnition 1.4. Let w w1,...,wn St n . We say that w is a k-Grassmannian = ( ) ∈ ( ) permutation, and we write w St n, k if: w1 1; w is evil-avoiding, i.e. w avoids the patterns 2413, 3214, 4132, and∈ 4213( ;) and w has= k recoils, that is, letters a in w such 1 that a 1 appears to the left of a in w. (Equivalently, w− has exactly k descents.) + SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 5

Definition 1.5. We associate to each w St n, k a sequence of partitions Ψ w 1 k ∈ ( ) 1 1 ( ) = λ ,...,λ as follows. Write the code (cf. Definition 2.3) of w− as c w− c ( ) 1 ( ) = = c1,...,cn ; since w− has k descents, c has k descents in positions we denote by ( ) i ai a1,...,ak. Set a0 0. For 1 i k, define λ n ai 0, , 0,cai 1 1,cai 1 2,...,cai . = ≤ ≤ = ( − ) −( ⋯ − + − + ) ´¹¹¹¹¸¹¹¹¹¶ai−1 See Table 3 for examples of the map Ψ w . ( ) Definition 1.6. Given a positive integer n and a partition λ properly contained in a length λ n length λ rectangle (we will later use the notation λ Val n ), ( ) × ( − ( )) ∈ ( ) we define an integer vector gn λ v1,...,vn of length n as follows. Write λ k1 kl ( ) = ( ) = µ1 , ,µl where ki 0 and µ1 µl. We assign values to the entries v1,...,vn (by performing⋯ ) the following> step> for ⋯ >i from 1 to l. ( )

(Step i) Set vn µi equal to µi. Moving to the left, assign the value µi to the ● − ﬁrst ki 1 unassigned components. ( − ) After performing Step l, we assign the value 0 to any entry vj which has not yet been given a value.

Remark 1.7. Note that in Step 1, we set vn µ1 , vn µ1 1, , vn µ1 k1 1 equal to µ1. − − − ⋯ − − + Example 1.8. We have g5 2, 1, 1 0, 1, 2, 1, 0 , g6 3, 2, 2, 1 0, 2, 3, 2, 1, 0 , g6 3, 1, 1 0, 0, 3, 1, 1, 0 . (( )) = ( ) (( )) = ( ) (( )) = ( ) The main result of this paper is Theorem 1.10. The definition of Schubert polynomial can be found in Section 2. Definition 1.9. We write a → b → c if the letters a, b, c appear in cyclic order in w. So for example, if w 1423, we have that 1 → 2 → 3 and 2 → 3 → 4, but it is not the case that 3 → 2 → 1 or= 4 → 3 → 2. We then define n 2 − (1.2) xyFact w x1 yn 1 k xi yn 1 k . ( ) = Mi 1 i 2Mk n ( − + − )⋯( − + − ) = i→+i≤1→≤ k + Theorem 1.10. Let w St n, k be a k-Grassmannian permutation, as in Definition 1.4, and write Ψ w λ1,∈ ,λ(k .) Then the (unnormalized) steady state probability is given by ( ) = ( ⋯ ) k

S −1 i (1.3) ψw xyFact w c gn λ , = ( ) Mi 1 ( ( )) = S −1 i where c gn λ is the double Schubert polynomial associated to the permutation with i ( ( )) code gn λ , and gn is given by Deﬁnition 1.6. ( ) In the case that each yi 0, Theorem 1.10 becomes Theorem 1.11 below. = Theorem 1.11. Let w St n, k , and let Ψ w λ1,...,λk . Let µ be the vector ∈ k( ) ( ) = ( ) n 1 n 2 2 i i Zn 2 µ ∶ 2− , 2− ,..., 2 − λ , where we view each partition λ as a vector in 0− , = ( ) i 1 ≥ ∑= adding trailing 0’s if necessary. Then when each yi 0, the unnormalized steady state = probability ψw is given by k µ S −1 i ψw x c gn λ , = Mi 1 ( ( )) = i S −1 i where c gn λ is the Schubert polynomial of the permutation with code gn λ . ( ( )) ( ) 6 DONGHYUNKIMANDLAURENK.WILLIAMS

Equivalently, writing λi λi ,λi ,... , we have that = ( 1 2 ) k µ x i i i ψw sλ Xn λ1 ,Xn λ2 ,... , = Mi 1 ( − − ) =

i i i where sλ Xn λ1 ,Xn λ2 ,... denotes the ﬂagged Schur polynomial associated to shape i ( − − ) i λ , where the semistandard tableaux entries in row j are bounded above by n − λj.

Remark 1.12. It follows from [AS16, Theorem 7.2] that if we set each yi 0, the n = partition function Zn w Sn ψw is equal to i 1 hn i x1, x2,...,xi 1, xi, xi , where = ∑ ∈ ∏ = − ( − n n) hi is the homogeneous symmetric polynomial. Note that Zn has i 0 i terms. ∏ = However, if the yi are not zero, then Zn does not factor. We illustrate Theorem 1.11 in Table 3 in the case that n 5. The quantity s w is deﬁned in Deﬁnition 5.8. = ( )

k w St 5,k Ψ w probability ψw s w ∈ ( ) ( ) 6,3,1 ( ) 0 12345 ∅ x( ) 0 5,2,0 ( ) 1 12354 1, 1, 1 x( )S13452 0 ( ) 4,1,0 ( ) 1 12435 2, 2, 1 x( )S14532 0 ( ) 4,1,1 ( ) 1 12453 2, 2 x( )S14523 0 ( ) 5,2,1 ( ) 1 12534 1, 1 x( )S12453 0 ( ) 3,1,1 ( ) 1 13245 3, 2 x( )S15423 0 ( ) 3,2,1 ( ) 1 13425 3, 1 x( )S15243 0 ( ) 3,3,1 ( ) 1 13452 3 x( )S15234 0 ( ) 4,2,0 ( ) 1 14235 2, 1, 1 x( )S13542 0 ( ) 4,2,1 ( ) 1 14253 2, 1 x( )S12543 0 ( ) 4,3,1 ( ) 1 14523 2 x( )S12534 0 ( ) 5,3,1 ( ) 1 15234 1 x( )S12354 0 ( ) 3,0,0 ( ) 2 12543 2, 2 , 1, 1, 1 x( )S14523S13452 0, −1 ( ) ( ) 2,0,0 ( ) 2 13254 3, 2 , 1, 1, 1 x( )S15423S13452 0, 0 ( ) ( ) 2,2,0 ( ) 2 13542 3 , 1, 1, 1 x( )S15234S13452 0, −1 ( ) ( ) 1,1,0 ( ) 2 14352 3 , 2, 2, 1 x( )S15234S14532 0, −1 ( ) ( ) 1,1,1 ( ) 2 14532 3 , 2, 2 x( )S15234S14523 0, −1 ( ) ( ) 2,2,1 ( ) 2 15342 3 , 1, 1 x( )S15234S12453 0, −1 ( ) ( ) 3,2,0 ( ) 2 15423 2 , 1, 1, 1 x( )S12534S13452 0, −2 ( ) ( ) ( ) 3 15432 3 , 2, 2 , 1, 1, 1 S15234S14523S13452 0, −1, −2 Table 3. Special( states) ( w) ( St 5,k) , the corresponding sequences( ) of ∈ ( ) partitions Ψ w , the steady state probabilities ψw, and vectors s w . ( ) ( )

Proposition 1.13. The number of evil-avoiding permutation in Sn satisﬁes the re- currence e 1 1, e 2 2, e n 4e n − 1 − 2e n − 2 for n 3, and is given by ( ) = ( ) = ( ) = ( ) ( ) ≥ n 1 n 1 2 + √2 − + 2 − √2 − (1.4) e n ( ) ( ) . ( ) = 2 This sequence begins as 1, 2, 6, 20, 68, 232, and occurs in [FI] as sequence A006012. The cardinalities St n, k also occur as sequence A331969. S ( )S SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 7

Remark 1.14. Let w n, h ∶ h, h − 1,..., 2, 1, h + 1, h + 2,...,n St n . In [Can16, Corollary 16], Cantini( gives) a= formula( for the steady state probability) ∈ of( state) w n, h , as a trivial factor times a product of certain (double) Schubert polynomials. Our( main) result generalizes this one. For example, for n 4 and n 5, [Can16, Corollary 16] gives a formula for the probabilities of states 1, 2=, 3, 4 , 1,=3, 4, 2 , 1, 4, 3, 2 , and the probabilities of states 1, 2, 3, 4, 5 , 1, 3, 4, 5, 2( , 1, 4, 5), 3(, 2 , and) 1(, 5, 4, 3, 2) . Mean- while, Theorem 1.11 gives( a formula) ( for all six) ( states when) n 4( (see Table) 1) and 20 of the 24 states when n 5. Using (1.4), we see that asymptotically= Theorem 1.11 = 2 √2 n−1 gives a formula for roughly ( + 2 ) out of the n − 1 ! states of St n . Note also that the Schubert polynomials that( occur) in the formulas( ) of [Can16] are all of the form Sσ a,n , where σ a, n denotes the permutation 1, a + 1, a + 2, . . . , n, 2, 3,...,n . However,( ) many of( the) Schubert polynomials arising( as (factors) of steady probabilities) are not of this form. Already we see for n 4 the Schubert = polynomials S1432 and S1243, which are not of this form. This paper is organized as follows. In Section 2 we provide background on partitions, permutations, and Schubert polynomials. In Section 3 we explore the combinatorics of evil-avoiding and k-Grassmannian permutations. In Section 4 we present Cantini’s results giving recursive formulas for z-deformed steady state probabilities of the inhomogeneous TASEP. In Section 5 we state Theorem 5.9, which says that for w St n, k , the z-deformed steady state probability ψw z is equal to a “trivial factor”∈ times( a) product of z-Schubert polynomials – certain polynomials( ) which reduce to double Schubert polynomials when z ∞ (as we prove in Section 6). Section 7 is devoted to the proof of Theorem 5.9,= which in turn implies Theorem 1.10 and Theorem 1.11. In Section 8 we recall Arita-Mallick’s formula for steady state proba- 1 bilities in terms of multiline queues (when yi 0); we then show that when w− is a Grassmann permutation, the multiline queues= of type w are in bijection with semistandard tableaux. In Section 9 we prove a multiline queue formula for z-deformed steady state probabilities (when yi 0), which generalizes Arita-Mallick’s result. Us- ing this formula we prove Theorem 1.3= (the monomial factor conjecture) in Section 10. Section 11 presents a few remarks on future directions, and Section 12 is an appendix containing some technical results which are used in the proof of Theorem 5.9. Acknowledgements: L.W. would like to thank Allen Knutson for interesting discussions, and would like to acknowledge the support of the National Science Foun- dation under agreements No. DMS-1854316 and No. DMS-1854512, as well as the support of the Radcliffe Institute for Advanced Study at Harvard University, where some of this work “took place” (virtually). Any opinions, findings and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.

2. Background on partitions, permutations and Schubert polynomials

We let Sn denote the symmetric group on n letters, which is a Coxeter group generated by the simple reﬂections s1,...,sn 1, where si is the simple transposition − exchanging i and i+1. We let w0 n, n−1,..., 2, 1 denote the longest permutation. = ( ) For 1 i n, we have the divided diﬀerence operator ∂i which acts on polynomials ≤ < P x1,...,xn as follows: ( ) P ...,xi, xi 1,... − P ...,xi 1, xi,... ∂iP x1,...,xn ( + ) ( + ). ( )( ) = xi − xi 1 + 8 DONGHYUNKIMANDLAURENK.WILLIAMS

If si1 ...sim is a reduced expression for a permutation w, then ∂i1 ...∂im depends only on w, so we denote this operator by ∂w.

Deﬁnition 2.1. Let x x1,...,xn and y y1,...,yn be two sets of variables, and let = ( ) = ( )

∆ x, y xi − yj . ( ) = i Mj n( ) + ≤ To each permutation w Sn we associate the double Schubert polynomial ∈ S 1 w x, y ∂w− w0 ∆ x, y , ( ) = ( ) where the divided diﬀerence operator acts on the x-variables.

Deﬁnition 2.2. A partition λ λ1,...,λr is a weakly decreasing sequence of positive integers. We set length λ =∶ (r and call) it the length of λ. We let mul λ be the ( ) = ( ) multiplicity of the largest part λ1 in λ, and let λlast denote the smallest part of λ.

For example, if λ 5, 5, 4, 3, 3 then length λ 5, mul λ 2, and λlast 3. = ( ) ( ) = ( ) = = Definition 2.3. Given a permutation w Sn, its (Lehmer) code is the vector c w ∈ ( ) = v1,...,vn where vi j i wj wi . If v1,...,vn is the code of a permutation, ( 1 ) = S{ > S < }S ( ) we let c− v1,...,vn denote the permutation with code v1,...,vn . The shape λ w of w is the( partition) obtained by sorting the components( of the co)de. ( ) Example 2.4. If w 1, 3, 5, 4, 2 , then c w 0, 1, 2, 1, 0 and λ w 2, 1, 1 . = ( ) ( ) = ( ) ( ) = ( ) The following is well-known. Zn Lemma 2.5. Given a vector c1, ⋯,cn 0, there exists a permutation w Sn such ( ) ∈ ≥ ∈ that c w c1, ⋯,cn if and only if ci + i n for all 1 i n. ( ) = ( ) ≤ ≤ ≤ Definition 2.6. We say that a permutation w is vexillary if and only if there does not exist a sequence i j k ℓ such that w j w i w ℓ w k . Such a permutation is also called< 2143< -avoiding.< ( ) < ( ) < ( ) < ( ) Definition 2.7. The flag of a vexillary permutation w is defined in terms of its code c w as follows. If ci w 0, let ei be the greatest integer j i such that cj w ci w . ( ) ( ) ≠ ≥ ( ) ≥ ( ) The flag φ w is then the sequence of integers ei, ordered to be increasing. ( ) Definition 2.8. Let Xi denote the family of indeterminates x1,...,xi. For d = d1,...,dn a weakly increasing sequence of n integers, we define SSYT λ,d to be (the set of semistandard) tableaux T with shape λ for which the entries in( the i)th row are bounded above by di. We also define the flagged Schur function type T sλ Xd1 ,...,Xdn x ( ). ( ) = T SSYTQ λ,d ∈ ( ) Remark 2.9. There are also flagged double Schur functions, which can be defined in terms of tableaux or via a Jacobi-Trudi type formula [Man01, Section 2.6.5]. Theorem 2.10. [Man01, Corollary 2.6.10] If w is a vexillary permutation with shape 1 λ w and with flags φ w f1,...,fm and φ w− g1,...,gm , then we have ( ) ( ) = ( ) ( ) = ( ) S w x; y sλ w Xf1 − Ygm ,...,Xfm − Yg1 , ( ) = ( )( ) i.e. the double Schubert polynomial of w is a flagged double Schur polynomial. SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 9

3. Combinatorics of evil-avoiding and k-Grassmannian permutations In this section we study the special states of our Markov chain whose probabilities are proportional to products of Schubert polynomials. Recall our definition of k- Grassmannian permutations St n, k from Definition 1.4. As we will see, the set St n, k is in bijection with a certain( ) set ParSeq n, k of sequences λ1,...,λk of k partitions.( ) Recall that our main result (see Theorem( 1.10)) states that( the probability) of each state in St n, k is proportional to a product of k Schubert polynomials, which are determined by( the) corresponding sequence of partitions. 1 Remark 3.1. Note that w contains a pattern p if and only if w− contains the pattern 1 1 p− . So w is evil-avoiding if and only if w− avoids 3142, 2431, 3241, and 3214. n We say that a sequence w1,w2,...,wn Z has a descent in position j if wj wj 1. ( ) ∈ > + Proposition 3.2. Let c c1,c2,...,cn be the code of w Sn. Then w avoids the = ( ) 1 ∈ patterns 3142, 3214, 2431, and 3241 (equivalently, w− is evil-avoiding) if and only if for each descent position j, if there is a b j such that: ≤ ● wb wb 1 ⋅⋅⋅ wj and 0 cb n − j, and b is maximal with these properties, < + < < < < then we must have cj 1 cj 2 ⋅⋅⋅ cj c 0. + = + = = + b = Remark 3.3. Let w w1,...,wn Sn. Note that having the bth entry of the code = ( ) ∈ c code w equal to cb means that there are precisely cb letters less than wb which = ( ) do not occur in positions 1 through b. The condition cj 1 cj 2 ⋅⋅⋅ cj c 0 means + = + = = + b = that these letters must occur in increasing order in positions j + 1,...,j + cb. 1 Remark 3.4. If w St n, 1 then code w− has a unique descent. This means that 1 ∈ ( ) ( ) w− is a Grassmannian permutation [Man01, Definition 2.2.3]. Equivalently, there is only one a such that a + 1 appears to the left of letter a in w. Conversely, if w is an inverse of a Grassmannian permutation that starts with 1, we have w St n, 1 . ∈ ( ) Proof of Proposition 3.2. We start by showing that if w fails to satisfy the condition of Proposition 3.2, then it must contain one of the patterns 3142, 3214, 2431, and 3241. If w fails to satisfy the condition of Proposition 3.2, then there is a descent position j and a b j with wb wb 1 ⋅⋅⋅ wj and 0 cb n − j, and b is maximal ≤ < + < < < < with this property, but we do not have cj 1 cj 2 ⋅⋅⋅ cj c 0. + = + = = + b = (1) We have wb wb 1 ⋅⋅⋅ wj wj 1. < + < < > + (2) By the definition of code, there must be cb letters C smaller than wb which appear to the right of wb; they must therefore appear to the right of wj. (3) Since cb n−j, there must be at least one letter wℓ wb which appears among positions< j, n . Let ℓ be minimal with this property.> [ ] (4) The fact that we do not have cj 1 cj 2 ⋅⋅⋅ cj c 0 implies that the letter + = + = = + b = wℓ appears to the left of some letter wr C, i.e. ℓ r. ∈ < Let us first consider the case that b j. If cb 1 0 then wb 1 is not the smallest letter = + ≠ + to appear in positions b + 1, n , so there exists some m b + 1 such that wb 1 wm. [ ] > + > Therefore the letters wb,wb 1,wm,wℓ give either an instance of the pattern 3214 or of the pattern 3241, depending{ + on whether} m ℓ or ℓ m. < < If b j and cb 1 0 then wb 1 is the smallest letter to appear in positions b + 1, n . = + = + [ ] But then the letters wb,wb 1,wℓ,wr form the pattern 3142. { + } If b j, the fact that b is maximal such that j + cb n implies that j + cb 1 n. < < + ≥ Since wb 1 ⋅⋅⋅ wj, the cb 1 letters which are less than wb 1 and to the right of + < < + + position b + 1 must lie in positions j + 1, n . But now since cb 1 n − j, all the letters [ ] + ≥ 10 DONGHYUNKIMANDLAURENK.WILLIAMS in positions j +1, n must be less than wb 1. In particular the letter wℓ defined in (3) [ ] + must be less than wb 1. But now the letters wb,wb 1,wℓ,wr form the pattern 2431. Therefore we have+ shown that if w avoids{ the patterns+ 3142} , 3214, 2431, and 3241, then it satisfies the condition of Proposition 3.2. In the other direction, suppose that w contains the pattern 3214. Let i k ℓ m < < < denote the positions of the letters of this pattern. Then wi wk implies that there > exists some j with i j k such that wi wi 1 ⋅⋅⋅ wj wj 1. We have ci 2 since ≤ < < + < < > + ≥ wk and wℓ are both less than wi. And all of the ci letters less than wi which occur to the right of wi must occur in positions j + 1, n . Moreover, the 4 in the pattern, [ ] representing the letter wm wi, occurs in a position in j + 1, n . Therefore ci n − j. > [ ] < Therefore there exists some b with i b j with wi wb wb 1 ⋅⋅⋅ wj and ≤ ≤ ≤ < + < < 0 cb n − j, and we choose b to be maximal with these properties. We have that wb < < is greater than both wk and wℓ. But then it is impossible for cj 1 cj 2 ⋅⋅⋅ cj c 0: + = + = = + b = by Remark 3.3 this means that the cb letters less than wb that appear to the right of wb must occur in increasing order in positions j + 1, j + 2,...,j + cb, but this is false since wk and wℓ (the 2 and 1 of the pattern) occur in the wrong order. Exactly the same argument holds if w contains the pattern 3241. Nearly the same argument holds if w contains the pattern 3142. Again let i k < < ℓ m denote the positions of the letters of this pattern. As before, since wi wk, we < > can find i j k such that wi wi 1 ⋅⋅⋅ wj wj 1, and we have 2 ci n − j. ≤ < < + < < > + ≤ < We then choose b maximal with i b j such that 0 cb n − j. But again using ≤ ≤ < < Remark 3.3 we see it is impossible to have cj 1 cj 2 ⋅⋅⋅ cj c 0 – the 142 in the + = + = = + b = pattern 3142 (i.e. the letters in positions k,ℓ,m) means that the letters less than wb do not occur in increasing order in consecutive positions. Finally, suppose that w contains the pattern 2431. Let h i ℓ m denote the < < < positions of the letters of this pattern. Since wi wℓ, there must be a descent position > j with j ℓ. Let j be minimal such that wh wh 1 ⋅⋅⋅ wj wj 1. Because wm wh, < < + < < > + < we know that ch 1. Moreover, the letters less than wh which are to the right of ≥ it must appear in positions j + 1, n . Because wℓ wh, we know that ch n − j. [ ] > < Therefore there exists some b h with wb wb 1 ⋅⋅⋅ wj and 0 cb n − j; choose ≥ < + < < < < b to be maximal with this property. But then by Remark 3.3, the cb letters less than wb which appear to the right of wb must appear in increasing order in positions j + 1, j + 2,...,j + cb. This is impossible, since wℓ and wm lie weakly to the right of position j + 1 but in the wrong order (since wℓ wm). This completes the proof. > Definition 3.5. We say that a partition λ is valid for n if λ is properly contained in a length λ × n − length λ rectangle. Let Val n λ λ is valid for n . ( ) ( ( )) ( ) = { S } n 1 Remark 3.6. It is easy to show that Val n 2 − − n − 1 − 1. The elements of S ( )S = ( ) Val n are in bijection with Grassmannian permutations in Sn that starts with 1. ( ) Definition 3.7. For 1 k n − 2, let ParSeq n, k denote the set of all sequences of partitions λ1,...,λk such≤ ≤ that each λi is valid( for) n, and for all 1 i k − 1: ( ) i ≤i 1≤ ● if ℓ is the smallest part of λ , then the first n − ℓ parts of λ + are equal. ( ) If k 0 then ParSeq n, k consists of one element, the empty sequence. = ( ) Example 3.8. If n 6, then 3 , 2, 2, 2, 1 and 4, 2 , 1, 1, 1, 1 lie in ParSeq 6, 2 but 3 , 2, 2, 1, 1 =and 4,((2 ,) 1(, 1, 1 do)) not lie(( in ParSeq) ( 6, 2)). ( ) (( ) ( )) (( ) ( ) ( ) Remark 3.9. It follows from Definition 3.7 that if λ1,...,λk ParSeq n, k , then for all 1 i k − 1: ( ) ∈ ( ) ≤ ≤ SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 11

i i 1 ● the number of parts of λ is less than the number of parts of λ + i i 1 ● every part of λ is greater than every part of λ + . ∶ n 1 n 2 2 Remark 3.10. Let v n denote the vector v n 2− , 2− ,..., 2 . Then it ( ) k i ( ) = ( ) follows from Definition 3.7 that µ ∶ v − i 1 λ is a partition. (Here we think of each i Zn 2 = ∑ = partition λ as a vector in 0− by adding extra parts equal to 0 if necessary.) Note that the vector µ appears in the≥ steady steady probability formula from Theorem 1.11. Proposition 3.11. The map Ψ ∶ St n, k → ParSeq n, k (cf. Definition 1.5) is well- ( ) 1 ( ) defined and bijective. The inverse map Ψ− ∶ ParSeq n, k → St n, k can be described 1 k ( ) ( ) as follows. Let λ ,...,λ ParSeq n, k , and let f1,...,fk be the sequence of first 1 ( k ) ∈ i ( )n f1 − 1 (+ n f2 −) 2 +⋅⋅⋅+ n fk − k parts of λ ,...,λ , i.e. fi λ1. Then f1 − λ f2 − λ fk − λ = 1 (( ) ( 1) 1 ( k )) is the code of a permutation w− of w St n, k . We define Ψ− λ ,...,λ w. ∈ ( ) ( ) = Proof. We first show that when we apply the map Ψ, we obtain a vector that sat- 1 isfies the properties of Definition 3.7. Write code w− c c1,...,cn and let ( ) = = ( j ) a a1,...,ak denote the positions of the descents of c. We have λ n − aj j − ⋯ ⋯ = ( + ) 0, , 0,caj−1 1, ,caj . If we take the maximal b such that aj 1 b aj and aj cb n, ( + ) ⋅⋅⋅ − < ≤ j <1 then by Proposition 3.2, we have caj 1 caj 2 caj c 0. Since λ + + + + b − aj+1 − ⋯ ⋯ = =+ = =j 1 = n aj 1 0, , 0,caj 1, ,caj 1 , t he first aj cb parts of λ + are equal. Let l ( + ) ( + + ) be the smallest part of λj, then l n − aj − cb. So the first n − l aj + cb parts of j 1 = ( ) ( ) = λ + are equal. 1 Now we show that when we apply the map Ψ− to λ1,...,λk ParSeq n, k , we n f1 − 1 + (n f2 − 2 +⋅⋅⋅+) ∈ n (fk − ) k get an element in St n, k . Let c f1 − λ f2 − λ fk − λ . ( ) i = (( ) ( ) ( )) Since the smallest part of λ is no greater than fi, the first n − fi parts of λi 1 are n fi+1 + equal. So the first n − fi components of the vector fi −1 − λi 1 are zero. We write ( + + ) c c1,...,cn as follows = ( ) 1 f1 − λ , if 1 j n − f1 ⎧ j ≤ ≤ ⎪ − i − − cj ⎪fi λj, if n fi 1 j n fi, for 2 i k = ⎨ − < ≤ ≤ ≤ ⎪ − ⎪0, if n fk j, ⎩⎪ < where we regard λi 0 if j is bigger that the length of λi. j = We claim that c c . If λi 0 then we have c f f f − n fi n fi 1 n fi n fi i i 1 i 1 − > − + − = − = > + ≥ + λi 1 c . If λi 0 then λi f since λi Val n . The first n − λi parts n+ fi 1 n fi 1 n fi n fi i n fi − + = − + − > − < ∈ ( ) − of λi 1 are equal so λi 1 f . Thus we have c f − λi 0 c . + n+ fi 1 i 1 n fi i n fi n fi 1 − + = + − = − > = − + We see that the descents of c are at n−f1,...,n−fk. Now take the maximal b such that n − fi 1 b n − fi and n − fi + cb n. Then cb fi − l where l is the smallest i − < ≤ − i 1 < = ⋅⋅⋅ part of λ . So the first n l parts of λ + are equal which implies cn fi 1 cn fi cb 1 − + = = − + as n − fi + cb n − l. We conclude that c is the code of w− for some w St n, k by Proposition 3.2.= ∈ ( )

1 1 Example 3.12. If code w− 0, 3, 1, 1, 0 then Ψ w 3 , 1, 1 . If code w− 0, 2, 2, 1, 0 then Ψ w ( 2),= 1(, 1, 1 . ) ( ) = (( ) ( )) ( ) = ( ) ( ) = (( ) ( )) Remark 3.13. While we have not found any previous works studying evil-avoiding permutations, we note that the sequence e n of cardinalities of evil-avoiding per- { ( )} mutations in Sn has several other combinatorial interpretations listed in [FI]:

● e n counts permutations π Sn for which the pairs i, π i with i π i , considered( ) as closed intervals∈ i + 1, π i , do not overlap;( ( )) equivalently,< ( for) each i n there is at most one[ j i with( )] π j i. ∈ [ ] ≤ ( ) > 12 DONGHYUNKIMANDLAURENK.WILLIAMS

● e n is the number of permutations on n with no subsequence abcd such that (i)( bc) are adjacent in position and (ii)[ max] a, c min b, d . For example, the 4 permutations of 4 not counted by a 4( are) <1324,(1423), 2314, 2413. ● e n is the number of rectangular[ ] permutations( ) on n , i.e. those permutations which( ) avoid the four patterns 2413, 2431, 4213, 4231[ ], see [CFF18]. It would be interesting to find a bijection between the k-Grassmannian permutations in Sn (where we let k vary) and any of the above sets of permutations. Proposition 3.14. For k 1, we have that ≥ n 1 ParSeq n, k 2 ParSeq n − 1,k + − ParSeq i, k − 1 . S ( )S = S ( )S i Qk 1 S ( )S = + Proof. To prove Proposition 3.14, we define two different injective maps Ψi ∶ ParSeq n− ( 1,k → ParSeq n, k for i 1, 2 as well as a family of injective maps Φi,k,n ∶ ParSeq i, k− 1 →) ParSeq n,( k )for k +=1 i n − 1. The statement then follows from the claim( that) every element( ) of ParSeq≤ n,≤ k lies in the image of precisely one of these maps. ( ) 1 k We define Ψ1 ∶ ParSeq n−1,k → ParSeq n, k to be the map which takes λ ,...,λ to µ1,...,µk , where µ(i is obtained) from (λi by) duplicating its first part.( That is, if) λi ( λi ,λi ,...,λ) i , then µi λi ,λi ,λi ,...,λi . So for example, = ( 1 2 r) = ( 1 1 2 r) Ψ1 2 , 1, 1 2, 2 , 1, 1, 1 . (( ) ( )) = (( ) ( )) 1 k We define Ψ2 ∶ ParSeq n−1,k → ParSeq n, k to be the map which takes λ ,...,λ to µ1,...,µk , where( for i ) 2, µi is obtained( ) from λi by duplicating( its first) ( ) i ≥i i i part. For i 1, let λ λ1,...,λr . If all parts of λ are equal, then we de- i i = i =i ( ) i i i i fine µ λ1 + 1,λ1,...,λr . Otherwise, we define µ λ1 + 1,λ2,...,λr . So if λ1 λ1=,λ(1,...,λ1 , then µ)1 λ1 + 1,λ1,λ1,...,λ1 . So= for( example, ) = ( 1 2 r) = ( 1 1 2 r) Ψ2 2 , 1, 1 3, 2 , 1, 1, 1 . (( ) ( )) = (( ) ( )) For k + 1 i n − 1, we define Φi,k,n ∶ ParSeq i, k − 1 → ParSeq n, k to be the ≤ ≤ 1 k 1 (k 1 ) j ( ) j map which takes λ ,...,λ to i − 1 ,µ ,...,µ − , where µ is obtained from λ ( ) ((j ) ) j j j j by duplicating the first part of λ n − i times. That is, if λ λ1,λ2,...,λr , then j j j j j j = ( ) µ λ ,λ ,...,λ ,λ ,...,λr . So for example, we have that = ( 1 1 1 2 ) n i 1 ´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹− + ¹¹¹¹¹¹¹¹¹¹¹¹¶ Φ4,2,5 1 3 , 1, 1 , (( )) = (( ) ( )) Φ4,2,5 2 3 , 2, 2, , (( )) = (( ) ( )) Φ4,2,5 1, 1 3 , 1, 1, 1 , (( )) = (( ) ( )) Φ4,2,5 2, 1 3 , 2, 2, 1 , and (( )) = (( ) ( )) Φ3,2,5 1 2 , 1, 1, 1 . (( )) = (( ) ( )) The above examples express the seven elements of ParSeq 5, 2 as images of elements of ParSeq 4, 2 , ParSeq 4, 1 , and ParSeq 3, 1 . ( ) ( ) ( ) ( ) Corollary 3.15. Define the number T n, k by ( ) n k 2 i + k − 1 n − 2 − i (3.1) T n, k − − 2i . ( ) = Qi 0 k − 1 k = (These numbers appear in [FI, A331969].) Then we have St n, k ParSeq n, k T n, k . S ( )S = S ( )S = ( ) SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 13

1 Equivalently, the number of evil-avoiding permutations w in Sn 1 such that w− has exactly k descents is T n, k . − ( ) Proof. The formula (3.1) is equivalent to the generating function given in A331969. By Proposition 3.14 and Proposition 3.11, it suﬃces to prove that n 1 T n, k 2T n − 1,k + − T i, k − 1 . ( ) = ( ) i Qk 1 ( ) = + We have n k 2 i + k − 1 n − 2 − i n k 3 i + k − 1 n − 3 − i T n, k − 2T n − 1,k − − 2i − 2 − − 2i ( ) ( ) = Qi 0 k − 1 k Qi 0 k − 1 k = = n − 2 n k 3 i + k n − 3 − i i + k − 1 n − 3 − i + 2 − − 2i − = k Qi 0 (k − 1 k k − 1 k ) = n k 3 n − 2 − − i 2 i + k − 1 n − 3 − i + 2 2 + , and so = k Qi 0 k − 2 k = T n, k − 2T n − 1,k − T n − 1,k − 2T n − 2,k ( ( ) ( )) ( ( ) ( )) − n k 3 + − − − − n k 4 + − − − n 2 − − i 1 i k 1 n 3 i n 3 − − i 1 i k 1 n 4 i + 2 + − + 2 + = ( k Qi 0 k − 2 k ) ( k Qi 0 k − 2 k ) = = − n k 4 + − − − − n 3 − − i 1 i k 1 n 4 i n k 2 n 4 + 2 + + 2 − − T n − 1,k − 1 . = k − 1 Qi 0 k − 2 k − 1 k − 2 = ( ) = Now the proof follows from the induction on n.

n 2 Proof of Proposition 1.13) Since e n − St n, k , by Proposition 3.14 we have ( ) = k 0 S ( )S ∑= n 2 n 2 n 1 e n − St n, k 1 + − 2 St n − 1,k + − St j, k − 1 ( ) = kQ0 S ( )S = kQ1( S ( )S j Qk 1 S ( )S) = = = + n 3 n 3 n 2 3 − St n − 1,k − 1 + − − St j, k − 1 = ( kQ0 S ( )S ) kQ1 j Qk 1 S ( )S = = = + n 3 n 3 n 2 4e n − 1 − − St n − 1,k + 1 − − − St j, k − 1 = ( ) (kQ0 S ( )S kQ1 j Qk 1 S ( )S) = = = + n 3 4e n − 1 − 1 + − 2 St n − 2,k 4e n − 1 − 2e n − 2 . = ( ) ( kQ1( S ( )S) = ( ) ( ) = 4. Cantini’s z-deformation of steady state probabilities In [Can16], Cantini associated to each w St n a deformed steady state probability ∈ ( ) ψw z ψw z1, ⋯, zn which recovers the usual steady state probability ψw for w in the( inhomogeneous) = ( TASEP) when “z ∞,” or in other words, when one reads off the = coefficient LCz ψw z of the largest monomial in z. In this section we define some operators on states( ( and)) polynomials, and then recall some results of [Can16]. Recall that indices of states are considered modulo n, so that for w w1,...,wn = ( ) ∈ St n , we have wn 1 w1. In this section we will also consider the indices for variables ( ) + = z z1,...,zn modulo n. = ( ) 14 DONGHYUNKIMANDLAURENK.WILLIAMS

Deﬁnition 4.1. For 1 i n − 1, the simple transposition si acts on the state ≤ ≤ w w1,...,wn St n by = ( ) ∈ ( )

si w1,...,wi,wi 1,...,wn w1,...,wi 1,wi,...,wn . ( + ) = ( + )

We also deﬁne the action of sn by

sn w1,w2,...,wn 1,wn wn,w2,...,wn 1,w1 . ( − ) = ( − ) And the shift operator σ acts on the state w by

σ w1,...,wn w2,...,wn,wn 1 . ( ) = ( + )

If f is a multivariate polynomial in z1,...,zn, then si acts on f z1,...,zn by permuting the variables, i.e. ( )

sif z1,...,zi, zi 1,...,zn f z1,...,zi 1, zi,...,zn . ( + ) = ( + )

And if z z1, z2, z3,... is an ordered set of variables, we let = ( )

σ z z2, z3, z4,... . ( ) = ( )

Finally we deﬁne LCz f to be the coeﬃcient of the highest degree term with respect to the z variables.( )

Proposition 4.2. [Can16, Equations (2), (24), (27), (28), (34), Theorem 18] Given w St n , we define ψw z ψw z1,...,zn recursively, via: ∈ ( ) ( ) = ( ) n i 1 n j i 1 − ψ 1,2, ,n z xi − yn 1 j − − zi − xj zi − yn 1 j , ( ⋯ )( ) = 1 Mi j n( + − ) Mi 1 Mj 1( ) jMi 1( + − ) ≤ < ≤ = = = + ψs w z πl wl,wl 1; n ψw z if wl wl 1, l ( ) = ( + ) ( ) > + where πl β,α; n is the isobaric divided difference operator defined by ( )

zl − yn 1 β zl 1 − xα G z − slG z πl β,α; n G z ( + − )( + ) ( ) ( ). ( ) ( ) = xα − yn 1 β zl − zl 1 + − + Then we have that for cyclically equivalent states w and σ w in St n , ( ) ( )

(4.1) ψσ w z1, ⋯, zn ψw σ z1, ⋯, zn , ( )( ) = ( ( )) where the indices on the z-variables are considered modulo n. Then ψw z is a polynomial in x, y, z, and its leading term LCz ψw z is ( ) ( ( ))

(4.2) LCz ψw z ψw. ( ( )) =

Because of (4.2), we refer to ψw z as the deformed steady state probability. ( ) n n n 1 n 4 Remark 4.3. For any w Sn, the total degree of ψw z is +n n−1 ( − )( + ) . ∈ ( ) 3 ( ) = 6

In what follows we will often omit n in the operator πl wl,wl 1; n and just write ( + ) πl wl,wl 1 when n is clear from the context. ( + ) SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 15

Example 4.4. For n 3, we have = ψ 1,2,3 z x1 − y1 z1 − y2 z1 − y1 z2 − x1 z2 − y1 z3 − x1 z3 − x2 , ( )( ) = ( )( )( )( )( )( )( ) ψ 3,2,1 z π3 3, 1 ψ 1,2,3 z z1 − x1 z2 − x1 z2 − y1 z3 − y1 × ( )( ) = ( ) ( )( ) = ( )( )( )( ) x1 + x2 − y1 − y2 z3z1 + x1x2 − y1y2 z3 + z1 − x1x2y1 − x1x2y2 + x1y1y2 + x2y1y2 , (( ) ( )( ) )) ψ 2,3,1 z π1 3, 2 ψ 3,2,1 z x1 − y1 z3 − y2 z3 − y1 z1 − x1 z1 − y1 z2 − x1 z2 − x2 , ( )( ) = ( ) ( )( ) = ( )( )( )( )( )( )( ) ψ 3,1,2 z π2 2, 1 ψ 3,2,1 z x1 − y1 z2 − y2 z2 − y1 z3 − x1 z3 − y1 z1 − x1 z1 − x2 , ( )( ) = ( ) ( )( ) = ( )( )( )( )( )( )( ) ψ 1,3,2 z π1 3, 1 ψ 3,1,2 z z2 − x1 z3 − x1 z3 − y1 z1 − y1 × ( )( ) = ( ) ( )( ) = ( )( )( )( ) x1 + x2 − y1 − y2 z1z2 + x1x2 − y1y2 z1 + z2 − x1x2y1 − x1x2y2 + x1y1y2 + x2y1y2 , (( ) ( )( ) )) ψ 2,1,3 z π2 3, 1 ψ 2,3,1 z z3 − x1 z1 − x1 z1 − y1 z2 − y1 × ( )( ) = ( ) ( )( ) = ( )( )( )( ) x1 + x2 − y1 − y2 z2z3 + x1x2 − y1y2 z2 + z3 − x1x2y1 − x1x2y2 + x1y1y2 + x2y1y2 . (( ) ( )( ) )) Taking the leading coeﬃcients with respect to z-variables recovers the probabilities from Figure 1, namely

ψ 1,2,3 ψ 2,3,1 ψ 3,1,2 x1 − y1 ( ) = ( ) = ( ) = ψ 1,3,2 ψ 2,1,3 ψ 3,2,1 x1 + x2 − y1 − y2. ( ) = ( ) = ( ) = 5. The z-deformation of our main result and z-Schubert polynomials In this section we present a z-generalization of our main result (see Theorem 5.9): for w St n, k , the z-deformed steady state probability ψw z is equal to a “trivial factor”∈ TF( w )times a product of z-Schubert polynomials (see( Definition) 5.3) – certain polynomials( in) x, y, z which reduce to double Schubert polynomials when z ∞. We note that the z-Schubert polynomials Sn z; x; y R z; x; y (where= z λ( ) ∈ [ ] = z1,...,zn , x x1,...,xn 1 , y y1,...,yn 1 ) are not defined for any permu- {tation but} rather= { depend on− a} choice= of{ positive− integer} n and a partition λ Val n . Given λ Val n , the polynomial Sn z; x; y has the property that: ∈ ( ) ∈ ( ) λ( ) ● when z ∞, this polynomial reduces to the double Schubert polynomial of = the permutation with code gn λ , see Proposition 5.7; ( ) ● when we additionally set each yi 0, we obtain a flagged Schur polynomial of shape λ, see Corollary 6.10. =

Definition 5.1. Given ordered variables x x1, x2,... , if I is a set of positive = ( ) integers we let xˆ denote xˆ x ∖ xi i I , where we keep the order on variables I I = { S ∈ } inherited from x. We abuse notation and use xî to denote x î . { }

Recall from Definition 4.1 that if z z1, z2,... then σ z denotes z2, z3,... . = ( ) ( ) ( ) Example 5.2. For f z; x; y x2z1z2 + z1 + z2 + x1 + x2 + x3, we have ( ) = 2 f σ z ; x; y x2z3z4 + z3 + z4 + x1 + x2 + x3 ( ( ) ) = f z; xˆ; y x3z1z2 + z1 + z2 + x2 + x3 + x4 ( 1 ) = f z; x 2,3 ; y x4z1z2 + z1 + z2 + x1 + x4 + x5. ( {Ã } ) = Definition 5.3. Given a positive integer n and a partition λ Val n (see Definition 3.5), we define the z-Schubert polynomial Sn z; x; y recursively∈ as follows:( ) λ( ) n 1 S − z; x; y 1, ∅ ( ) = 16 DONGHYUNKIMANDLAURENK.WILLIAMS

∅ Sn and for λ , we define λ z; x; y to be (5.1) ≠ ( ) n mul λ λ1 λ2 1 n λ1 mul λ 1 − ( ) ( − + ) − − ( )+ ⋯ Sn 1 λ1 λ2 1 − − ∂n λ1 mul λ ∂1 λ′− σ − + z; x1ˆ; y x1 yl zi xm , − − ( ) ( ) Ml 1 ( ) Mi 1 mM2 ( ) = = = where the divided difference operators act on the x-variables, and where λ′ is the partition obtained by deleting the first part of λ. (If λ′ is empty we regard λ2 0.) = Remark 5.4. Using induction on n, it is easy to show that the only z-variables Sn ⋯ appearing in λ z; x; y are z1, , zλ1 length λ . ( ) + ( ) Remark 5.5. For positive integers r and s, Cantini introduced a polynomial that he Sr,s Sr s 1 denoted in [Can16]. In our notation this is the same as +r −1 s−1 z; x; y . (( − ) )( ) Example 5.6. We have S3 − − − − 1 z; x; y ∂1 x1 y1 x1 y2 z1 x2 z2 x2 ( )( ) = ( )( )( )( ) x1 − y1 x1 − y2 z1 − x2 z2 − x2 − x2 − y1 x2 − y2 z1 − x1 z2 − x1 ( )( )( )( ) ( )( )( )( ) = x1 − x2 z1z2 x1 + x2 − y1 − y2 − z1 + z2 x1x2 + y1y2 + x1x2 y1 + y2 − y1y2 x1 + x2 . = ( ) ( )( ) ( ( ) ( )) S3 + − − Observe that LCz 1 z; x; y x1 x2 y1 y2, the double Schubert polynomial ( ( )( )) = S 1,3,2 x, y . It also appears as a unnormalized steady state probability in Figure 1. ( )( ) Recall the definition of gn λ from Definition 1.6. ( ) Proposition 5.7. Fix n and choose a partition λ Val n . Then the leading coeffi- n n S S ∈ ( ) S −1 cient LCz λ z; x; y of λ z; x; y is the double Schubert polynomial c gn λ of ( ( 1 )) ( ) ( ( )) the permutation c− gn λ Sn whose code is gn λ . ( ( )) ∈ ( ) We will prove Proposition 5.7 in the next section. In Theorem 5.9 below we give a z-deformation of our main result (Theorem 1.10); it says that for w St n, k , the z-deformed steady state probability ψw z is equal to a “trivial factor”∈ TF( w )times a product of z-Schubert polynomials. ( ) Recall that we defined( xyFact) w in Definition 1.9. We also define ( n) (5.2) xzFact w zi − xj ( ) = Mi 1 jMwi ( ) = ≠ min wi,wi 1,wi 2,...,j j { − − }= and n (5.3) yzFact w zi − yj . ( ) = Mi 1 j nM1 wi ( ) = ≠ + − max wi,wi 1,wi 2,...,n 1 j n 1 j { + + + − }= + − Finally we set (5.4) TF w xyFact w xzFact w yzFact w . ( ) = ( ) ( ) ( ) 1 k We associate to a partition sequence λ ,...,λ a vector a1,...,ak as follows. ( ) ( ) Definition 5.8. Let λ1,...,λk be a sequence of partitions. We denote the parts j j j ( ) 1 k Zk of λ by λ1,λ2, etc. We define s λ ,...,λ ; n a1,...,ak by setting a1 0, and for each 2 i k, (( ) ) = ( ) ∈ = ≤ ≤ i 1 i 1 ai ai 1 + λ1− + length λ − − n. = − ( ) SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 17

j j + 1 j j + 1 i − m i − m

i i

Figure 2. The ﬁgure shows the ladder move Li,j.

If w St n, k such that Ψ w λ1,...,λk , then abusing notation, we also refer to s λ1∈,...,λ( k ); n as s w . ( ) = ( ) (( ) ) ( ) See Table 3 for examples of s w . The following theorem will be proved in Section 7. ( ) 1 k Theorem 5.9. Let w St n, k , and write Ψ w λ , ⋯,λ and s w a1, ⋯, ak . Then we have ∈ ( ) ( ) = ( ) ( ) = ( ) k Sn ai (5.5) ψw z TF w λi σ z ; x; y ( ) = ( ) Mi 1 ( ( ) ) = where subscripts for z variables are considered modular n.

6. Properties of z-Schubert polynomials 6.1. Double Schubert polynomials. We review an algorithmic formula for com- puting double Schubert polynomials in terms of rc-graphs, based on [BB93]. Definition 6.1. Given a finite subset D 1, 2, ⋯ × 1, 2, ⋯ we define the weight of D to be ⊊ { } { } wt D x, y xi − yj . ( )( ) = i,jMD( ) ( )∈ The initial diagram of the permutation w is

Din w i, j 1 j c w i . ( ) = {( ) S ≤ ≤ ( ) } Definition 6.2. For a finite subset D 1, 2, ⋯ × 1, 2, ⋯ , assume the following conditions are satisfied for some i and j:⊊ { } { } ● i, j D, i, j + 1 D, ● (i −)m,∈ j ,(i − m, j) +∉ 1 D for some 0 m i, ● (i − k, j ), (i − k, j + 1 ) ∉D for each 1 k< m<. ( ) ( ) ∉ ≤ < Then we define the ladder move Li,j to be Li,j D D ∪ i − m, j + 1 ∖ i, j . We represent diagrams D as above as collections( of)+=’s, see{( Figure 2. We)} also{( define)} L D to be the set of all D′ that can be obtained by applying ladder moves to D. ( ) Billey-Bergeron [BB93, Theorem 3.7] showed that L Din w gives the set of rc- graphs for w. Combining this with the formula of Fomin-Kirillov( ( )) [FK96] produces the following formula for Schubert polynomials (see also [KM05, Corollary 2.1.5]). Theorem 6.3. [FK96], [BB93, Theorem 3.7 and Lemma 3.2] Let w be a permutation.

(a) We have Sw x, y wt D′ x, y . ′ ( ) = D ∑Din w ( )( ) ∈L( ( )) 18 DONGHYUNKIMANDLAURENK.WILLIAMS

t (b) The map sending D′ to its transpose D′ is a bijection between L Din w 1 ( ) ( ( )) and L Din w− . ( ( )) Example 6.4. The diagrams L Din 1, 4, 2, 3 are shown in Figure 3. So we have ( ( )) S 1,4,2,3 x, y x2 − y1 x2 − y2 + x2 − y1 x1 − y3 + x1 − y2 x1 − y3 . ( )( ) = ( )( ) ( )( ) ( )( )

1 2 3 1 2 3 1 2 3 1 1 1 2 2 2

Din 1, 4, 2, 3 (( ))

Figure 3. The three diagrams in L Din 1, 4, 2, 3 . ( ( )) Note that rc-graphs are in bijection with reduced pipe dreams (replace each + with a crossing of two wires and each empty position with a pair of “elbows”). We will need the following result about linear factors of certain double Schubert polynomials. If yi 0 for all i, Proposition 6.5 follows from [BB93, Corollary 3.11]. = Proposition 6.5. Let w Sn be a permutation, and let c w c1,...,cn and 1 ∈ 1 ( ) = ( ) c w− c˜1,..., c˜n denote the codes of w and w− . Suppose there is some l 0 such ( ) = ( ) ≥ that c˜m 0 for all m l, and let w′ be the permutation with code c w′ l,c1,...,cn (whose existence= follows> from Lemma 2.5). Then ( ) = ( )

l S S w′ x, y w x1ˆ, y x1 − yk . ( ) = ( ) Mk 1( ) =

Remark 6.6. The condition that c˜m 0 for all m l is equivalent to the condition that w has an increasing subsequence= of the form l +> 1,l + 2,...,n.

Proof. We first claim that no D′ L Din w contains a + in a column greater than ∈ ( ( )) l, i.e. there is no i, j D′ with j l. If there were such a D′, then we’d have t ( ) ∈ 1 > j, i D′ . But Din w− does not have an element whose x-coordinate is bigger (than) l∈,( and) ladder moves( never) increase the x-coordinates of the +’s involved. By the claim, Din w does not contain a + in a column greater than l, which ( ) implies the same is true for Din w′ and hence for L Din w′ . ( ) ( ( )) Now we define a map f ∶ L Din w →L Din w′ by ( ( )) ( ( )) f D′ i + 1, j i, j D′ ∪ 1, 1 , 1, 2 , ⋯, 1,l . ( ) = {( ) S ( ) ∈ } {( ) ( ) ( )} This map is clearly injective, and is well defined since f Din w Din w′ . We ( ( )) = ( ) claim that f is surjective. Assume not. Then we can find D1,D2 L Din w′ such ∈ ( ( ) that D2 Li,j D1 , and D1 is in the image of f but D2 is not. = ( ) Clearly the only way that there would be a viable ladder move Lij on D1 which 1 does not have a counterpart for f − D1 is if Lij adds a + in row 1, necessarily in ( ) some column j l (since the first component of c w′ is l). But we know that no > ( ) diagram in L Din w′ can have a + in a column greater than l. Therefore the map f is surjective( and( hence)) bijective. SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 19

We conclude that

Sw′ x, y wt D′′ x, y wt f D′ x, y ′′ ′ ′ ( ) = D QDin w ( )( ) = D QDin w ( ( ))( ) ∈L( ( )) ∈L( ( )) l x1 − yk wt D′ x1ˆ, y ′ = D QDin w Mk 1( ) ( )( ) ∈L( ( )) = l l S x1 − yk wt D′ x1ˆ, y x1 − yk w x1ˆ, y . ′ = Mk 1( ) D QDin w ( )( ) = Mk 1( ) ( ) = ∈L( ( )) =

6.2. The proof of Proposition 5.7. The following lemma is easy to verify.

Lemma 6.7. Let w Sn with code c w v1, ⋯, vn . Then wi wi 1 if and only if ∈ ( ) = ( ) > + vi vi 1. In this case we have > + (6.1) ∂iSw Sws , = i ⋯ − + and c wsi v1′ , , vn′ , where vi′ vi 1, vi′ 1 vi 1, and vj′ vj for j i, i 1 . ( ) = ( ) = + + = = ∉ { } If we iterate (6.1), we ﬁnd that if v1 − i vi 1 for all 1 i r, then ≥ + ≤ ≤ S −1 S −1 (6.2) ∂r∂r 1 ...∂1 c v1,...,vn x, y c v2,v3,...,vr+1,v1 r,vr+2,...,vn x, y . − ( )( ) = ( − )( ) We are now ready to prove Proposition 5.7. Proof of Proposition 5.7. We use induction on the number of parts of λ. If λ ℓ has one part, then the deﬁnition of z-Schubert polynomial plus (6.2) gives = ( )

n 1 n − S ⋯ − ⋯ S 1 LCz ℓ z; x; y ∂n ℓ 1 ∂1 x1 yl ∂n ℓ 1 ∂2∂1 c− n 1,0, ,0 x, y ( ( )( )) = − − (Ml 1 ( )) = − − ( ( − ⋯ )( )) = ∂n ℓ 1⋯∂2 Sc−1 0,n 2,0, ,0 x, y Sc−1 0, ,0,ℓ,0, ,0 x, y = − − ( − ⋯ )( ) = ( ⋯ ⋯ )( ) S −1 c gn ℓ x, y . = ( (( )))( ) Now let λ λ1, ⋯,λk with k 1 and assume the statement holds for partitions = ( ) > with at most k − 1 parts. Setting λ′ λ2, ⋯,λk , we have = ( ) n mul λ − ( ) Sn Sn 1 λ1 λ2 1 LCz λ z; x; y ∂n λ1 mul λ ⋯∂1 LCz λ′− σ − + z; x1ˆ; y x1 − yl ( ( )) = − − ( ) ( ( )) Ml 1 ( ) = n mul λ − ( ) ⋯ S −1 ′ − ∂n λ1 mul λ ∂1 c gn−1 λ x1ˆ, y x1 yl . = − − ( ) ( ( ))( ) Ml 1 ( ) = 1 Let gn 1 λ′ v1, ⋯, vn 1 and let w c− v1,...,vn 1 Sn 1. From Remark 1.7 we − ( ) = ( − ) = ( − ) ∈ − have vi λ′ for n − 1 − λ′ i n − mul λ′ − λ′ . This plus Deﬁnition 1.6 implies that = 1 1 ≤ ≤ ( ) 1 wn mul λ′ λ′ n − mul λ′ − ( )− 1 = ( ) ⋮

wn 2 λ′ n − 2 − − 1 = wn 1 λ′ n − 1. − − 1 = 1 It follows from Remark 6.6 that if we write c w− c˜1,..., c˜n 1 , then c˜m 0 for ( ) = ( − ) = all m n − mul λ′ − 1. Since n − mul λ n − mul λ′ − 1, we have c˜m 0 for all > ( ) ( ) ≥ ( ) = 20 DONGHYUNKIMANDLAURENK.WILLIAMS m n − mul λ , and Proposition 6.5 implies that > ( ) n mul λ − ( ) S −1 − S −1 c v1,...,vn−1 x1ˆ, y x1 yl c n mul λ ,v1, ,vn−1 x, y . ( )( ) Ml 1 ( ) = ( − ( ) ⋯ )( ) = Therefore we have that n S ⋯ S −1 LCz λ z; x; y ∂n λ1 mul λ ∂1 c n mul λ ,v1, ,vn−1 x, y . ( ( )) = − − ( ) ( − ( ) ⋯ )( ) We will apply (6.2), but need to ﬁrst check that n − mul λ1 vi + i for 1 i ( ) ≥ ≤ ≤ n − λ1 − mul λ . To see this, note that since v1,...,vn 1 gn 1 λ′ , each vi λ1. ( ) ( − ) = − ( ) ≤ But then vi + i λ1 + n − λ1 − mul λ n − mul λ , as desired. Applying (6.2) gives ≤ ( ) = ( ) n S S −1 S −1 LCz λ z; x; y c v1, ,v mul ,λ1,v mul 1, ,vn−1 x, y c gn λ x, y . ( ( )) = ( ⋯ n−λ1− (λ) n−λ1− (λ)+ ⋯ )( ) = ( ( ))( )

6.3. z-Schubert polynomials and flagged Schur polynomials. We relate z- Schubert polynomials Sn z; x; y with flagged Schur polynomials (Corollary 6.10). λ( ) We start with Lemma 6.8 which is immediate from the definition of gn λ . ( ) Lemma 6.8. Let n be a positive integer and λ be a partition of length at most n − 2.

● The first and last components of gn λ are always zero. ( ) ● If we order the entries of gn λ in weakly decreasing order, we obtain λ. ( ) Lemma 6.9. Let λ be a partition of length n − 2 , and write λ µk1 ,...,µkl . ≤ ( ) = ( 1 l ) Let w be the permutation whose code is gn λ . Then w is vexillary, and if we write 1 ( ) k1 kl ⋅⋅⋅ the flag of w and w− (cf. Definition 2.7) as f1 ,...,fl (where f1 fl) and b1 b ( ) ≤ ≤ g ,...,g l , respectively, we have that ( 1 l ) f1,gl , f2,gl 1 ,..., fl,g1 n−µ1, n−k1 , n−µ2, n−k1−k2 ,..., n−µl, n−k1−⋅ ⋅ ⋅−kl . {( ) ( − ) ( )} = {( ) ( ) ( )} Proof. Recall that the essential set of a diagram D w is the set of southeast corners of the diagram, that is, ( ) Ess w i, j D w i + 1, j , i, j + 1 , i + 1, j + 1 D w . ( ) = {( ) ∈ ( ) S ( ) ( ) ( ) ∉ ( )} Writing λ µk1 , ⋯,µkl as in Definition 1.6, we see that the essential set of w is = ( 1 l ) Ess w n − µ1, n − k1 , n − µ2, n − k1 − k2 , n − µ3, n − k1 − k2 − k3 ,... . ( ) = {( ) ( ) ( ) } This essential set lies on a piecewise linear curve always oriented SW-NE, so by [Man01, Proposition 2.2.8], the permutation w must be vexillary. The claim about the flags follows immediately from the definition of the flag in terms of the code, or alternatively from [Man01, Exercise 2.2.11].

k1 k2 kl n Corollary 6.10. Write λ µ ,µ ,...,µ . Then LCz S z; x; y equals the = ( 1 2 l ) ( λ( )) ﬂagged double Schur polynomial sλ Xn µ1 − Yn k1 ,Xn µ2 − Yn k1 k2 ,...,Xn µ − Yn k1−⋅⋅⋅−k . ( − − − − − − l − l ) ´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹k1 ¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶ ´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹k2 ¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶ ´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸kl ¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶ n And when we set yi 0 for all i, LCz S z; x; y equals the ﬂagged Schur polyno- = ( λ( )) mial sλ Xn−λ1 ,Xn−λ2 ,... . ( ) Proof. This follows from Lemma 6.9, Proposition 5.7 and Theorem 2.10.

Example 6.11. The permutation 13542 has code 0, 1, 2, 1, 0 g5 2, 1, 1 . There- fore for λ 2, 1, 1 , we have that ( ) = (( )) = ( ) S5 S − − − LCz 2,1,1 z; x; y 13542 x, y s 2,1,1 X3 Y4,X4 Y2,X4 Y2 . ( ( )( )) = ( ) = ( )( ) SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 21

7. Proof of Theorem 5.9 In this section we prove Theorem 5.9, which in turn implies Theorem 1.10 and Theorem 1.11. Our strategy will be to prove Theorem 5.9 first in the case of w St n, 1 , and then use induction on k to prove it for w St n, k . ∈ (We note) that in this section, the divided difference operator∈ ( s act) on the z-variables. Sn a For brevity we will often denote the z-Schubert polynomial λ σ z ; x; y with Sn a ( ( ) ) shifted z-variables by λ σ z . As in the previous section, the subscripts for z- variables are considered modulo( ( ))n. Definition 7.1. For a partition λ Val n , we identify it with the lattice path L λ; n cutting out the Young diagram∈ that( takes) unit steps south and east from the ( ) upper right corner λ1, n − λ1 to the lower left corner 0, 0 of the rectangle. Label ( ) ( ) the vertical steps from the top to bottom with numbers 1 through n − λ1. Then label the horizontal steps from the right to the left with numbers n − λ1 + 1 through n. We define w λ; n to be the permutation of length n obtained by reading off the numbers along the( lattice) path. See Figure 4 for an example. Clearly w λ; n St n, 1 . As we will see in Proposition 7.3 (1), Ψ w λ; n λ . ( ) ∈ ( ) ( ( )) = ( )

9 8 2 3

11 10 4 5

13 12 6 7

Figure 4. For λ 6, 6, 4, 4, 2, 2 Val 13 , we can read oﬀ w λ;13 1, 2, 8, 9, 3, 4, 10, 11=,(5, 6, 12, 13, 7) from∈ ( the) lattice path L λ;13( cutting) = (out the Young diagram. ) ( )

Our first goal is to analyze the trivial factor TF w for w w λ; n (Proposition 7.5). This will help us prove Theorem 5.9 in the case( of )w St =n, 1(. We) start by refining the quantities introduced in (1.2), (5.2), and (5.3). ∈ ( ) Definition 7.2. Fix a positive integer n and choose w St n . We define ∈ ( ) xyFact w; i x1 − yn+1−k ⋯ xi − yn+1−k for 1 i n − 2 ( ) = i+1Mk n ( ) ( ) ≤ ≤ i→i+<1→≤ k

xzFact w; i zi − xj for 1 i n ( ) = jMwi ( ) ≤ ≤ ≠ min wi,wi 1,wi 2,...,j j { − − }= yzFact w; i zi − yj for 1 i n. ( ) = j nM+1−wi ( ) ≤ ≤ ≠ max wi,wi 1,wi 2,...,n+1−j n+1−j { + + }= 22 DONGHYUNKIMANDLAURENK.WILLIAMS

Clearly we have that n−2 n n xyFact w xyFact w; i , xzFact w xzFact w; i , yzFact w yzFact w; i . ( ) = Mi 1 ( ) ( ) = Mi 1 ( ) ( ) = Mi 1 ( ) = = = Proposition 7.3. Let λ Val n and w w1,...,wn w λ; n . Recall that w St n, 1 . The following statements∈ ( ) hold: = ( ) = ( ) ∈ ( ) −1 n−λ1 (1) We have that Ψ w λ; n λ . Equivalently, c w λ1 − λ, where we regard the vectors( ( on the)) = right-hand( ) side as vectors( ) of= length n by adding trailing 0’s. (2) Suppose that wi lies on a vertical step of L λ; n . Let A be the set of numbers ( ) on the horizontal steps below wi and B be the set of numbers on the vertical steps that are on the same vertical line as wi and below wi. We have

wi−1 xzFact w; i zi − xk ( ) = Mk 1 ( ) = yzFact w; i zi − yn+1−k . ( ) = k MA∪B( ) ∈ (3) Suppose wi lies on a horizontal step of L λ; n . Let C be the set of numbers ( ) on the vertical steps above wi and D be the set of numbers on the horizontal steps that are on the same horizontal line of wi and to the right of wi. We have

xzFact w; i zi − xk ( ) = k MC∪D( ) ∈ n yzFact w; i zi − yn+1−k . ( ) = k Mwi+1( ) = −1 Proof. (1) The numbers n−λ1 +1 through n appear in increasing order in w so c w ( ) vanishes after the n − λ1 st component. For 1 k n − λ1, let wi be the letter on ( ) ≤ ≤ the kth vertical step of L λ; n . Then there are λ1 − λk numbers bigger than wi in ( ) w1, ⋯,wi−1 (where we regard λk 0 if k length λ . Thus the kth component of −1 −1 = > ( ) c w is λ1 − λk, and c w has a unique descent in position n − λ1. The fact that Ψ( w λ); n λ now follows( from) the deﬁnition of Ψ. ( ( )) = (2) The numbers 1 through wi appear in increasing order in w so we have xzFact w; i wi−1 ( ) = zi − xk . To compute yzFact w; i , we need to ﬁnd all letters ℓ which are max- k 1 ( ) ( ) ∏= imum among wi,wi+1,...,ℓ and for each one we pick up a factor of zi − yn+1−ℓ . Clearly these letters{ are precisely} the ones in A ∪ B. ( ) (3) The proof is similar to part (2). Example 7.4. Let λ 6, 6, 4, 4, 2, 2 Val 13 as in Figure 4. We have = ( ) ∈ ( ) w w λ;13 1, 2, 8, 9, 3, 4, 10, 11, 5, 6, 12, 13, 7 = ( ) = ( ) c w−1 0, 0, 2, 2, 4, 4, 6, 0, 0, 0, 0, 0, 0 ( ) = ( ) λn−λ1 − λ 0, 0, 2, 2, 4, 4, 6 . 1 = ( ) For w5 3, we have = xzFact w;5 z5 − x1 z5 − x2 ( ) = ( )( ) yzFact w;5 z5 − y10 z5 − y1 z5 − y2 z5 − y3 z5 − y4 . ( ) = ( )( )( )( )( ) SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 23

b ˜ ˜ Proposition 7.5. Let λ Val n with mul λ b. Write λ λ1 , λ for some λ. b ˜ ∈ (′ ) b−(1 ) = ˜ = (( ) ) For w w λ1 , λ ; n and w w λ1 ,λ1 − 1, λ ; n , the following statements are true.= ((( ) ) ) = ((( ) ) ) ′ (1) If b 1, we have w sbw and wb wb+1. > =′ < (2) If b 1, we have σ w s1w and w1 w2. (3) If b = 1, we have ( ) = < > mul λ1−1,λ˜ −1 (( )) TF sbw M1 xb − yλ1 zb+1 − yn+1−b−i ( ) = ( ) Mi 1 ( ) = b−1 b+λ1−λ˜1

TF w M1 zb − yλ1 zb+1 − xb zi − yn+1−b zi − xn+1−λ1 . ( ) = ( )( ) Mi 1( ) iMb+2 ( ) = = for some rational expression M1 which is symmetric in variables zb and zb+1. (4) If b 1 , we have = n−λ1 mul λ1−1,λ˜ −1 (( )) TF s1w M2 x1 − yλ1 z1 − xi z2 − yn−i ( ) = ( ) Mi 1 ( ) Mi 1 ( ) = = 1+λ1−λ˜1

TF w M2 z1 − yλ1 z2 − x1 zi − xn+1−λ1 . ( ) = ( )( ) Mi 3 ( ) = for some rational expression M2 which is symmetric in variables z1 and z2. Proof. Part (1) and part (2) are straightforward from Deﬁnition 7.1. b ˜ (3) For 1 i b − 1, wb b is on the same vertical line with wi in L λ1 , λ ; n ′ ≤ ≤ = b−1 −((( ˜ ) ) ) but wb+1 b is not on the same vertical line with wi in L λ1 ,λ1 1, λ ; n . By Proposition= 7.3 (2), whenever 1 i b − 1, we have ((( ) ) ) ≤ ≤ ′ ′ (7.1) xzFact w; i yzFact w; i xzFact w ; i yzFact w ; i zi − yn+1−b . ( ) ( ) = ( ) ( )( ) ˜ For b + 2 i b + λ1 − λ1, wb+1 n + 1 − λ1 is on the same horizontal line with wi ≤b ˜≤ ′ + =− in L λ1 , λ ; n but wb n 1 λ1 is not on the same horizontal line with wi in ((( b−1) ) )˜ = L λ1 ,λ1 − 1, λ ; n . By Proposition 7.3 (3), whenever 1 i b − 1, we have ((( ) ) ) ′≤ ≤ ′ (7.2) xzFact w; i yzFact w; i zi − xn+1−λ1 xzFact w ; i yzFact w ; i . ( ) ( )( ) = ( ) ( ) ′ ′ ′ Note that w + ,w + , ⋯,w are on the same vertical line if and only if λ1 − 1 b 1 b 2 b+mul λ˜ +1 = ˜ ( ) λ1. By Proposition 7.3 (2), we have (7.3)

b−1 λ1 xzFact w; b yzFact w; b zb − xk zb − yk ( ) ( ) = Mk 1( ) Mk 1( ) = = b−1 λ1−1 mul λ1−1,λ˜ −1 ′ ′ (( )) xzFact w ; b + 1 yzFact w ; b + 1 zb+1 − xk zb+1 − yk zb+1 − yn+1−b−k . ( ) ( ) = Mk 1( ) Mk 1 ( ) Mk 1 ( ) = = = And by Proposition 7.3 (3), we have

b λ1−1 (7.4) xzFact w; b + 1 yzFact w; b + 1 zb+1 − xk zb+1 − yk ( ) ( ) = Mk 1( ) Mk 1 ( ) = = b−1 λ1−1 ′ ′ xzFact w ; b yzFact w ; b zb − xk zb − yk . ( ) ( ) = Mk 1( ) Mk 1 ( ) = = 24 DONGHYUNKIMANDLAURENK.WILLIAMS

′ ′ For w, we have b − 1 → b → wb+1 n + 1 − λ1 while for w we have b → b + 1 → w = b = n + 1 − λ1. So we conclude

b−1 xk − yλ1 xyFact w k∏1( ) 1 (7.5) = . ( ′) b xyFact w = = xb − yλ1 ( ) xk − yλ1 k 1( ) ∏= Combining (7.1), (7.2), (7.3),(7.4) and (7.5) proves the argument. (4) The proof is similar to part (3).

Proposition 7.6. Theorem 5.9 is true for w St n, 1 . ∈ ( ) Proof. We use induction on λ for Ψ w λ . The base case λ 0 corresponds to the identity permutation in StS Sn, 0 . Take( ) = any( )w St n, 1 suchS thatS = Ψ w λ . Let (b ˜ ) ˜ ∈ ( )′ b−(1 ) = ( ˜) b mul λ and write λ λ1 , λ for some λ. Denoting w w λ1 ,λ1−1, λ ; n , by= the( induction) hypothesis= (( we) have) = ((( ) ) )

′ n ψ ′ z TF w S 1 z . w λ1 b− ,λ1−1,λ˜ ( ) = ( ) (( ) )( ) We ﬁrst prove the induction step when b 1. By Proposition 7.5 (1) and Proposition 4.2, we have > (7.6) − − zb yλ1 zb+1 xb ′ n ψ z π w + ,w ψ ′ z ∂ TF w S 1 z . w b b 1 b w ( )( ) b λ1 b− ,λ1−1,λ˜ ( ) = ( ) ( ) = xb − yλ1 ( ) (( ) )( ) By Proposition 7.5 (3) we can write

mul λ1−1,λ˜ −1 ′ (( )) TF w M1 xb − yλ1 zb+1 − yn+1−b−i ( ) = ( ) Mi 1 ( ) = b−1 b+λ1−λ˜1

TF w M1 zb − yλ1 zb+1 − xb zi − yn+1−b zi − xn+1−λ1 ( ) = ( )( ) Mi 1( ) iMb+2 ( ) = = for some M1 that is symmetric in variables zb and zb+1. Plugging this into (7.6) gives

mul λ1−1,λ˜ −1 n (( )) ψ z M1 z − y 1 z +1 − x ∂ ⎛S 1 z z +1 − y +1− − ⎞ w b λ b b b λ1 b− ,λ1−1,λ˜ b n b i ( ) = ( )( ) (( ) )( ) Mi 1 ( ) ⎝ = ⎠ By Proposition 12.5 we have

b−1 b+λ1−λ˜1 n ψ z M z − y 1 z + − x ⎛S z z − y + − z − x + − 1 ⎞ w 1 b λ b 1 b λ1 b,λ˜ i n 1 b i n 1 λ ( ) = ( )( ) (( ) )( ) Mi 1( ) iMb+2 ( ) ⎝ = = ⎠ TF w Sn z . = ( ) λ( ) Now consider the case b 1. By Proposition 7.5 (2) and Proposition 4.2, we have (7.7) = − − z1 yλ1 z2 x1 ′ n ψ z π w ,w ψ ′ z ∂ TF σ w S σ z . w 1 2 1 σ w ( )( ) 1 λ1−1,λ˜ ( ) = ( ) ( )( ) = x1 − yλ1 ( ( )) ( )( ( )) SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 25

By Proposition 7.5 (4) we can write

n−λ1 mul λ1−1,λ˜ −1 ′ (( )) TF σ w M2 x1 − yλ1 z1 − xi z2 − yn−i ( ( )) = ( ) Mi 1 ( ) Mi 1 ( ) = = 1+λ1−λ˜1

TF w M2 z1 − yλ1 z2 − x1 zi − xn+1−λ1 . ( ) = ( )( ) Mi 3 ( ) = for some M2 that is symmetric in variables z1 and z2. Plugging this into (7.7) gives

n−λ1 mul λ1−1,λ˜ −1 n (( )) ψ z M2 z1 −y 1 z2 −x1 ∂1 ⎛S σ z z1 − x z2 − y − ⎞ w λ λ1−1,λ˜ i n i ( ) = ( )( ) ( )( ( )) Mi 1 ( ) Mi 1 ( ) ⎝ = = ⎠ By Proposition 12.4 we have

1+λ1−λ˜1 n ψ z M z − y 1 z − x ⎛S z z − x + − 1 ⎞ w 2 1 λ 2 1 λ1,λ˜ i n 1 λ ( ) = ( )( ) ( )( ) Mi 3 ( ) ⎝ = ⎠ TF w Sn z . = ( ) λ( )

Deﬁnition 7.7. If π Sm and σ Sp, the direct sum π ⊕ σ Sm+p is the permutation ∈ ∈ ∈ π i if 1 i m deﬁned by π ⊕ σ i ( ) ≤ ≤ ( )( ) = σ i − m + m if m + 1 i m + n. For example, 3, 2, 1 ⊕ (3, 1, 2,)5, 4 3, 2, 1, 6,≤4, 5≤, 8, 7 . ( ) ( ) = ( ) The following lemma is easy to verify. Lemma 7.8. Given λ Val n , let ∈ ( ) length λ +λ1−n u u1,...,un ∶ σ ( ) w λ; n . = ( ) = ( ( )) Let w¯ λ; n ∶ u1,...,un−λlast . Then w¯ λ; n Sn−λlast and u w¯ λ; n ⊕idλlast , where ( ) = ( ) ( ) ∈ = ( ) idm is the identity permutation on m letters.

Example 7.9. Let n 5 and λ 2, 2 Val 5 , so that λlast 2. Then w λ;5 = = ( ) ∈ ( ) −1 = ( ) = 1, 2, 4, 5, 3 and length λ + λ1 − n −1. We have u σ 1, 2, 4, 5, 3 3, 1, 2, 4, 5 , ( ) ( ) = = ( ) = ( ) so w¯ λ;5 3, 1, 2 . We have u 3, 1, 2 ⊕ id2. ( ) = ( ) = ( ) Proposition 7.10. Let w St n, k for k 2 with Ψ w λ1,λ2, ⋯,λk and s w ∈ ( a)2 ≥ 1 ( ′) = ( ′ ) ( ↓) = a1,...,ak . Then we can write σ w w¯ λ ; n ⊕ w for some w . If we let w ∶ ( )′ ↓ ( ) = ↓( )2 k = id − 1 ⊕w , then w St n, k − 1 and Ψ w λ , ⋯,λ . n λlast ∈ ( ) ( ) = ( ) Example 7.11. Let w 1, 2, 5, 4, 3 St 5, 2 . Then Ψ w λ1,λ2 2, 2 , 1, 1, 1 = ( ) ∈ ( ) ( ) = ( ) = (( ) ( )) and s w a1, a2 0, −1 . From the previous example, w¯ λ;5 3, 1, 2 . We have a2 ( ) = ( ) = ( ) ′ ′ ( ) = (↓ ) ′ σ w 3, 1, 2, 5, 4 3, 1, 2 ⊕w where w 2, 1 . And we have w idn−λ1 ⊕w ( ) = ( ) = ( ) = ( ) = last = 1, 2, 3 ⊕ 2, 1 1, 2, 3, 5, 4 St 5, 1 with Ψ w↓ λ2 1, 1, 1 . ( ) ( ) = ( ) ∈ ( ) ( ) = ( ) = (( )) 1 2 1 1 −1 1 n−λ1 1 2 n−λ1 Proof. Note that a2 length λ +λ1 −n. We have c w λ1 −λ + λ1 − k 2 k n−λ1 = k ( ) ( ) = (( ) ) (( ) λ + ⋯ + λ1 − λ and by the definition of ParSeq n, k we know that the first − i )− 1 (( ) i )n λ1 − i ( ) − 1 n λlast parts of λ1 λ are zero for 2 i k. So the first n λlast parts of 1 ( −1 ) 1 ((n−λ1 ) 1 ) ≤ ≤ 1 −1 ( ) c w equal λ1 − λ which is the same as c w λ ; n by Proposition 7.3 ( ) (( ) ) ((−( 1 )) ) (1). So the positions of the numbers 1 through n λlast are the same in w and 1 1 1 1 ( length) λ +λ1−n 1 w λ ; n . Since taking the first n − λ parts of σ ( ) w λ ; n gives ( ) ( last) ( ( )) 26 DONGHYUNKIMANDLAURENK.WILLIAMS

1 + 1− − 1 length λ λ1 n w¯ λ; n , we conclude taking the first n λlast parts of σ ( ) w also gives w¯(λ; n). ( ) ( ) ( ) −1 Note that the i-th component of c w counts the number of wj i for 1 1 + 1− −1 −1 length( λ )λ1 n > − 1 ≤ j w i . Thus c w and c σ ( ) w coincide after the n λlast -th ≤ ( ) ( ) ( ( ) − 1 ( ) component as taking the numbers smaller than n λlast to the front does not 1 ( ) ′ −1 n − λ c id 1 ⊕w affect the code of its inverse after last -th component. Thus n−λlast −1 (− 1 ) (( − 1 ) ) coincides with c w after the n λlast component and the first n λlast parts ( ) ( ) ( )′ −1 id 1 c id 1 ⊕w are zero as the permutation starts with n−λlast . We conclude n−λlast − 2 − k (( ) ) = λ2 n λ1 − λ2 + ⋯ + λk n λ1 − λk and we are done. (( 1) ) (( 1) ) ′ ′ Lemma 7.12. Let u and u be permutations in Sn and w and w be permutations in Sm. We have TF u′ ⊕ w TF u′ ⊕ w′ ( ) ( ). TF u ⊕ w = TF u ⊕ w′ ( ) ( ) Proof. It is enough to show the following three equations xyFact u′ ⊕ w xyFact u′ ⊕ w′ (7.8) ( ) ( ) xyFact u ⊕ w = xyFact u ⊕ w′ ( ) ( ) xzFact u′ ⊕ w xzFact u′ ⊕ w′ (7.9) ( ) ( ) xzFact u ⊕ w = xzFact u ⊕ w′ ( ) ( ) yzFact u′ ⊕ w yzFact u′ ⊕ w′ (7.10) ( ) ( ). yzFact u ⊕ w = yzFact u ⊕ w′ ( ) ( ) For 1 i n−1 we have xyFact u⊕w; i xyFact u⊕w′; i and xyFact u′ ⊕w; i xyFact u≤′ ⊕≤w′; i . For n+1 i n(+m−1 )we= have xyFact( u⊕)w; i xyFact( u′ ⊕w); i= and xyFact( u ⊕)w′; i xyFact≤ ≤u′ ⊕ w′; i . And for i n(we have) =xyFact u(⊕ w; i ) xyFact u′ ⊕(w; i and)xyFact= u(⊕ w′; i xyFact) u′ ⊕ w=′; i . So the first equation( (7.8)) = follows.( ) ( ) = ( ) For 1 i n we have xzFact u ⊕ w; i xzFact u ⊕ w′; i and xzFact u′ ⊕ w; i xzFact u≤′⊕w≤′; i . Now consider the( case n) +=1 i n(+m. For)1 j n, xzFact( u⊕w); i= ( ) ′ ≤ ≤ ≤ ≤ ( ) has a factor zi − xj for xzFact u ⊕ w ; i has a factor zi − xj . And the same is true for xzFact u(′ ⊕ w; i )and xzFact( u′ ⊕ w′;)i . For n + 1 (j n +)m xzFact u ⊕ w; i has ( ) ′ ( ) ≤ ≤ ( ) a factor zi − xj for xzFact u ⊕ w; i has a factor zi − xj . And the same is true for xzFact u(⊕ w′; i) and xzFact( u′ ⊕ w′); i . So the second( equation) (7.9) follows. The( proof for) (7.10) is similar( to the) proof for (7.9).

Proposition 7.13. [Can16, Theorem 20] Let u Sn and w Sm. Then we can write ∈ ∈ 1 2 ψu⊕w z ψ z ψ z ( ) = u( ) w( ) where ψ1 z (respectively ψ2 z ) depends only on u (respectively w).4 u( ) w( ) ′ ′ Corollary 7.14. Let u and u be permutations in Sn and w and w be permutations in Sm. We have ψu′⊕w z ψu′⊕w′ z ( ) ( ). ψu⊕w z = ψu⊕w′ z ( ) ( ) 4 The result stated in [Can16, Theorem 20] concerns the skew sum of the permutations u and w, not the direct sum; however, the direct sum of u and w is a cyclic rotation of the skew sum of w and u, so the result we’ve stated follows. SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 27

Proof of Theorem 5.9. We prove Theorem 5.9 for w St n, k using induction on k. Theorem 5.9 holds for k 1 by Proposition 7.6. Suppose∈ ( the theorem) holds for all elements in St n, k − 1 . = ( ) 1 2 k We now consider w St n, k . Let Ψ w λ ,λ , ⋯,λ and s w a1, a2, ⋯, ak . By Proposition 7.10, we∈ have( )σa2 w ( w¯) =λ1(; n ⊕ w′ for) some w( ′.) Moreover,= ( if we) ↓ ′ ( )↓= ( ) ↓ 2 k w ∶ id 1 ⊕w w St n, k − 1 Ψ w λ , ⋯,λ set n−λlast , we have that and . Using = ∈↓ ( ) ( ) = ( ) Deﬁnition 5.8 it is easy to see that s w 0, a3 − a2, ⋯, ak − a2 . By the induction hypothesis we have ( ) = ( )

k ′ − 2 1 n ai a (7.11) ψ 1 ⊕ ′ z TF id − ⊕w S i σ z . idn−λ w n λlast λ last ( ) = ( ) Mi 2 ( ( )) = 1 1 1 By Lemma 7.8, and using the fact that a2 length λ +λ1−n, we have that w¯ λ ; n ⊕ a2 1 1 = ( ) ( ) id 1 σ w λ ; n , where w λ ; n St n, 1 . Therefore the induction hypothesis λlast = ( ( )) ( ) ∈ ( ) and (4.1) implies that

1 n a1−a2 ψ 1 ⊕ z TF w¯ λ ; n ⊕ id 1 S 1 σ z . (7.12) w¯ λ ;n id 1 λlast λ ( ) λlast ( ) = ( ( ) ) ( ( )) By Corollary 7.14 we have

1 ψ ⊕ 1 z ψ 1 ⊕ ′ z w¯ λ ;n idλ idλ w ( ) last ( ) n−last ( ) ψw¯ λ1;n ⊕w′ z ( ) ( ) = ψid 1 ⊕ id 1 z n−λlast λlast ( ) Plugging this into (7.11) and (7.12) and using Lemma 7.12 gives

1 ψ ⊕ 1 z ψ 1 ⊕ ′ z w¯ λ ;n idλ idλ w ( ) last ( ) n−last ( ) ψw¯ λ1;n ⊕w′ z ( ) ( ) = ψid 1 ⊕ id 1 z n−λlast λlast ( ) ′ 1 1 1 k TF w¯ λ ; n ⊕ idλ TF idλ ⊕w last n−last Sn ai−a2 ( ( ) ) ( ) λi σ z = TF idn−λ1 ⊕ idλ1 Mi 1 ( ( )) ( last last ) = k 1 ⊕ ′ Sn ai−a2 TF w¯ λ ; n w λi σ z . = ( ( ) ) Mi 1 ( ( )) = Now using (4.1) and cyclically shifting z-variables completes the proof.

8. Multiline queues and steady state probabilities

When each yi 0, there is a combinatorial formula for the steady state probabilities of the inhomogeneous= TASEP in terms of the multiline queues of Ferrari and Martin [FM07]. This result (Theorem 8.4) was conjectured in [AL14] and proved in [AM13]. In this section we show in Theorem 8.5 that if w−1 is a Grassmann permutation, then the multiline queues of type w are in bijection with certain collections of nonintersecting paths, which in turn are in bijection with semistandard tableaux. We note that in the case that w w0,s1w0 or s2w0, multiline queues were related to nonintersecting paths in [AL18, Section= 3]. And in the case that all particles have types 0, 1, 2, nonintersecting paths were used to give explicit determinantal formulas for steady state probabilities in [Man17, Theorem 2.6]. Multiline queues were also connected to tableaux via nonintersecting paths in [AGS20], though the weights on queues there were diﬀerent from ours. 28 DONGHYUNKIMANDLAURENK.WILLIAMS

Definition 8.1. Fix positive integers L and n. A multiline queue Q of content m m1,...,mL is an L × n array in which there are m1 +⋅⋅⋅+ mi balls in row i for 1 =i( L. We label) the rows 1,...,L from top to bottom. ≤Given≤ such an array, there is a bully path labeling procedure which assigns a label to each ball. We start by assigning each ball in row 1 the label 1. Then we consider the leftmost of these balls (call it b) and construct its 1-bully path: match b to the first unmatched ball b′ in row 2 which is encountered when one looks directly below b then travels right in row 2 (wrapping around if necessary); then match b′ to the first ball b′′ in row 3 which is encountered when one looks below b′ then travels right as before. Continuing, we obtain the 1-bully path for b; we label all matched balls by 1. We then construct 1-bully paths for each other ball in row 1 (considering them from left to right). We then assign the label 2 to all unmatched balls in row 2 (considered left to right), and similarly construct their 2-bully paths. Continuing in this way gives the bully path labeling to all balls in Q, see Figure 5.

1 2 1 2 2 3 2 1 2 2

Figure 5. A multiline queue of type 2, 2, 1, 4, 4, 4, 2, 3 . ( )

After completing the bully path projection for Q, let w w1, ⋯,wn be the labeling of the balls read from right to the left in row L (where a vacancy= ( is denoted) by L+1). This will be a composition (not necessarily a permutation). We say that Q is a multiline queue of type w and let MLQ w denote the set of multiline queues of type w. We also deﬁne the type of row r in Q( to) be the labeling of the balls in row r read from right to left (where a vacancy is denoted by r + 1). A vacancy in Q is called i − covered if it is traversed by an i-bully path, but not by an i′-bully path for i′ i. A trivial bully path is a bully path that goes straight down from its starting point.<

Deﬁnition 8.2. Given an L × n multiline queue Q, let vr be the number of vacancies L in row r and zr,i be the number of i − covered vacancies in row r. Let Vi vj. = j ∑i+1 The weight of the multiline queue Q is deﬁned by =

L−1 zr,i Vi xr wt Q xi . ( ) = Mi 1 1 Mi r L xi = ≤ < ≤ Example 8.3. The multiline queue Q in Figure 5 is of type 2, 2, 1, 4, 4, 4, 2, 3 . Its ﬁrst and second rows have types 2, 2, 2, 2, 2, 1, 2, 2 and 3, 3(, 2, 2, 1, 3, 2, 3 , respec-) tively. There is one 1-covered vacancy( in the second) row and( two 1-covered) vacancies in the third row. There is no 2-covered vacancy. The weight of Q is

4+3 3 x2 x3 2 4 4 2 wt Q x1 x2 x1x2x3. ( ) = (x1 )(x1 ) = The next theorem was conjectured in [AL14] and proved in [AM13]. It holds for the inhomogeneous TASEP on a ring where yi 0 for all i, and the weights of particles can be any positive numbers (with repeats allowed).= SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 29

Theorem 8.4. [AM13] Let w w1,...,wn be a composition, and consider the inho- = ( ) mogeneous TASEP on a ring (with yi 0 for all i) whose states are all compositions obtained by permuting the parts of w.= of w. Then the (unnormalized) steady state probability ψw can be expressed as a weight-generating functions for multiline queues, that is,

ψw wt Q . = Q MLQQ w ( ) ∈ ( ) Theorem 8.5 below says that there is a bijection between multiline queues associated to Grassmann permutations and flagged semistandard tableau (see Definition 2.8). After giving some preparatory lemmas, we state the bijection in Definition 8.11 and illustrate it in Example 8.12. Note that Theorem 8.5 gives a new proof of Theorem 1.11 when k 1. = Theorem 8.5. Given λ Val n , let d n − λ1, n − λ2, ⋯, n − λlength λ . Then there is a bijection f ∶ MLQ w∈ λ;(n ) → SSYT= ( λ,d such that the number( )) of i-covered vacancies in row r of Q( MLQ( ))w λ; n equals( ) the number of r’s in row i of f Q . ∈ ( ( )) ( ) × r r Lemma 8.6. Given an L n multiline queue Q, let w( ) wi( ) be the type of row r = { } r in Q, where the subscript i in wi( ) refers to the ith column of Q, read right to left. Suppose that there exist a b L′ L and i, j in 1, 2,...,n such that: < ≤ ≤ { } L w( ) a i = r ′ w( ) b, for L r L j = ≤ ≤ r ′ w( ) a, for L − 1 r L and k i + 1, ⋯, j − 1 k ≠ ≤ ≤ ∈ { } + ⋯ − + ⋯ ⋯ − r where i 1, , j 1 denotes i 1, , n, 1, , j 1 if j i. Then wi( ) a for L′ − 1 {r L. See Figure} 6. { } < = ≤ ≤ Proof. By assumption the statement holds for r L. We prove it for r L − 1, L − ′ − r = =′ 2,...,L 1 in that order. Assume we have wi( ) a for some r between L and L. If r−1 r−1 ⋯ =− wi( ) a then we have wk( ) a for k in i, , j 1 . So the a-bully path going to ≠ r ≠ r{ } r the ball wi( ) a traverses the position wj( ) b. Since b r, the position wj( ) is not a vacancy. It= is a contradiction since the a-bully= path traverses≤ the ball labeled by r−1 b a. Thus we have w( ) a. > i =

No balls with label a in this area

row L′ − 1 a row L′ b b a b b a row L b ⋯ a b ⋯ a column j column i column j column i

Figure 6. The diagrams at the left and right of this ﬁgure illustrate the hypothesis and conclusion of Lemma 8.6. 30 DONGHYUNKIMANDLAURENK.WILLIAMS

Recall the deﬁnition of w λ; n from Deﬁnition 7.1. Corollary 8.7 shows that the bottom part of many of the( bully) paths in Q MLQ w λ; n will be trivial (a column of balls with the same label, as in Figure∈ 7), which( ( in turn)) will help us show that the corresponding tableau we associate to Q has the correct number of rows.

r r Corollary 8.7. For λ Val n and Q MLQ w λ; n , let w( ) wi( ) be the type ∈ ( ) ∈ r ( ( )) = { } of row r in Q, where the subscript i in wi( ) refers to the ith column, read right to left. Choose any 1 i n. n−1 ≤ ≤ r n−1 ( ) ( ) ( ) − 1 − If 1 wi length λ , then wi wi for n λw(n− ) r n 1. In other ≤ ≤ ( ) = i ≤ ≤ words, the height of the trivial part of the bully path in column i will be λ (n−1) . wi + n−1 r n−1 n−1 − If length λ 1 wi( ) n, then wi( ) wi( ) for wi( ) r n 1. In other ( ) ≤ ≤ = ≤ ≤ − n−1 words, the height of the trivial part of the bully path in column i will be n wi( ). n−1 ( ) − 1 Proof. Note that if wi n, n 1 or λw(n− ) 1 then the claim is vacuous. There = i = are several cases to consider. -Case 1: Suppose 1 i λ1 + length λ − 1. We use induction on i starting from ≤ ≤ ( ) the largest number to the smallest. The base case i λ1 + length λ − 1 is trivial n−1 = ( ) ( ) − 1 since we have either wi n 1 or λw(n− ) 1. Suppose the desired statement = i = + − n−1 holds for j such that i j λ1 length λ 1. If 1 wi( ) length λ then we n−1 < ≤ ( )n−1 n−≤1 ≤ ( ) ( ) ( ) ( ) + 1 1 have either 1 wi+1 length λ with wi+1 wi 1 and λw(n− ) λw(n− ) or ≤ ≤ ( ) = i = i+1 n−1 n−1 + ( ) − ( ) − 1 + length λ 1 wi+1 n 1 with wi+1 n λw(n− ) 1. Either way, the induction ( ) ≤ ≤ = i r n−1 ( ) ( ) − 1 + − hypothesis implies that wi+1 wi+1 for n λw(n− ) 1 r n 1, and applying = i ≤ ≤ + n−1 − Lemma 8.6 gives the claim for i. If length λ 1 wi( ) n 1 then consider k n−1 n−1 + + ( ) ≤ ≤ such that wk( ) wi( ) 1. If k i 1 then the claim follows from Lemma 8.6 as = =n−1 n−1 + ( ) − 1 + ( ) before. If k i 1 we have 1 wl length λ and n λw(n− ) 1 wk for > ≤ ≤ ( ) l = + − r n−1 n−1 − i 1 l k 1. By the induction hypothesis, we have wl( ) wl( ) for wi( ) r n 1 ≤+ ≤ − r n−1 n−1 = − + ≤ ≤ − and i 1 l k 1. In particular, wl( ) wi( ) for wi( ) r n 1 and i 1 l k 1. Now applying≤ ≤ Lemma 8.6 gives the claim≠ for i. ≤ ≤ ≤ ≤ Case 2: If i λ1 +length λ , the desired statement is trivial. And if λ1 +length λ + 1 i n, then= the proof of( the) statement is similar to the proof of Case 1. ( ) ≤ ≤ Example 8.8. Consider λ 3, 3, 2, 1 Val 9 and w λ;9 1, 2, 7, 3, 8, 4, 9, 5, 6 . For Q MLQ w λ;9 , Corollary= ( 8.7) implies∈ ( that) the( bottom) = ( portions of many of) the i-bully∈ paths( ( are)) trivial (end with a sequence of vertical steps), as shown in Figure 7.

5 6 5 2 1 6 5 3 7 2 1 6 5 4 8 3 7 2 1

Figure 7. Part of a multiline queue in MLQ w 3, 3, 2, 1 ;9 MLQ 1, 2, 7, 3, 8, 4, 9, 5, 6 . ( ( ) ) = ( ) SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 31

Lemma 8.9. Fix λ Val n and Q MLQ w λ; n , and consider row r of Q. If ∈ ( ) ∈ ( ( )) n − λ1 r n − 1 then the numbers 1 through n − λ1 occur in increasing order as we ≤ ≤ r read the row type w( ) (from right to left as usual). And if r n − λ1, the numbers r < 1 through r occur in increasing order in w( ). It follows that the i-bully paths for 1 i n − λ1 are non-intersecting and they do not wrap around. ≤ ≤ Proof. We use induction on r starting from the largest number to the smallest. The base case r n − 1 is trivial. = First consider the case n−λ1 r n−1 and assume the statement holds for r +1 . ≤ < r+1 ( ) Then by the induction hypothesis, wa(i ) i for i n − λ1 and a1 ⋯ an−λ1 . = r ≤ ≤ < < For 1 i n − λ1 − 1, it is immediate that w( ) i for some ai bi ai+1, otherwise ≤ ≤ bi = ≤ < the i-bully path must cross over the ball labeled i + 1 in row r + 1 and column ai+1, r which is a contradiction. It remains to show w( ) n − λ1 for some an−λ1 bn−λ1 . bn−λ1 r = r+1 ≤ ( ) ( ) − + By Corollary 8.7 we have wk k for 1 k mul λ and wmul λ+1 n λ1 1. Thus = ≤ ≤ ( ) ( ) = bn−λ1 mul λ + 1 and if an−λ1 bn−λ1 then the n − λ1 -bully path must cross over ≥ ( ) > ( ) the ball labeled n − λ1 + 1 in row r + 1. For the case r n − λ1, we use induction on r, treating row n − λ1 as our base case, and decreasing< r by 1 at each step. Assume the statement holds for row r + 1 . ⋯ (r+1 ) Then by the induction hypothesis we have positions a1 ar+1 such that wa(i ) i length( ) λ +1 ( ) = length + 1 ( ( ) ) length + 1 λ so by Corollary 8.7 we know wλ1+length λ +1 λ . Since ( ) ( ) = ( ) length λ -bully path cannot traverse the position λ1 + length λ + 1, length λ + 1 , ( ) ( ( ) ( ) ) it must pass λ1 + length λ , length λ . By Lemma( 8.9, i-bully( paths) for 1( ))i length λ are non-intersecting and length λ - ≤ ≤ ( ) ( ) bully path passes the point λ1 +length λ , length λ . So length λ −1 -bully path ( ( ) ( )) ( ( ) ) must pass the point λ1 + length λ − 1, length λ − 1. Repeating this we deduce ( ( ) ( )) each i-bully path must pass the position λ1 + i, i for 1 i length λ . From the n−1 ( ) ≤ ≤ ( ) deﬁnition of w λ; n we have w( + −) i and by Corollary 8.7 i-bully path only moves ( ) λ1 i λi = vertically from the position λ1 +i−λi, n−λi . So it is enough to assign a lattice path ( ) from λ1 + i, i to λ1 + i − λi, n − λi to decide i-bully path. Note( that i)-bully( paths for i length) λ are trivial bully paths by Corollary 8.7. So we can identify Q MLQ w> λ; n (with) a set of non-intersecting lattice paths ∈ ( ( )) {P1, ⋯,Plength λ where Pi is a path from λ1 + i, i to λ1 + i − λi, n − λi . ( )} ( ) ( ) Now we can deﬁne the bijection whose existence was asserted in Theorem 8.5.

Deﬁnition 8.11. Given λ Val n , set d n − λ1, n − λ2, ⋯, n − λlength λ , and choose some Q MLQ w λ;∈n .( We) associate= ( to Q a set of non-intersecting( )) lattice ∈ ( ( )) 32 DONGHYUNKIMANDLAURENK.WILLIAMS paths P1, ⋯,Plength λ as in Lemma 8.10. We then let f Q denote the semistandard tableaux{ with length( )λ} rows, where the entries in the ith( row) record the row numbers ( ) of the horizontal steps of path Pi. Clearly the ith row has λi entries which are bounded by n − λi, so f Q SSYT λ,d . ( ) ∈ ( )

1 2 1

3 2 1 1 1 4 6 ≤ 4 3 2 1 2 3 6 6 5 4 3 2 1 ≤ 6 5 4 3 2 1 3 4 7 ≤ 6 5 4 3 7 2 1 6 8 ≤ 6 5 4 8 3 7 2 1

Figure 8. The left ﬁgure shows the multiline queue Q in MLQ w λ;9 for λ 3, 3, 2, 1 Val 9 . And the right ﬁgure shows the corresponding( ( )) semistandard= ( ) young∈ ( tableau.)

Example 8.12. The left of Figure 8 shows a multiline queue Q in MLQ w λ;9 for λ 3, 3, 2, 1 Val 9 . As in Lemma 8.10 we associate a set of non-intersecting( ( )) = ( ) ∈ ( ) lattice paths P1,P2,P3,P4 from positions λ1 + i, i to positions λ1 + i − λi, n − λi (indicated with{ green boxes).} ( ) ( ) The figure on the right shows the corresponding semistandard Young tableau obtained by the map above. Note that the 1-bully path has horizontal steps in rows 1, 1 and 4, so we fill the first row of λ with 1, 1 and 4. Proof of Theorem 8.5. By Lemma 8.10, we can identify the bully paths of each multiline queue Q MLQ w λ; n with a collection of non-intersecting lattice paths ∈ ( ( )) P1, ⋯,Plength λ , where Pi is a lattice path from λ1 +i, i to λ1 −λi +i, n−λi ; these {lattice paths( in)} turn get mapped via Definition 8.11( to) a semis( tandard tableaux) in SSYT λ,d , where d n − λ1, n − λ2, ⋯, n − λlength λ . Clearly the non-intersecting condition( on) lattice paths= ( corresponds to the strictly( incr)) easing condition on columns of the corresponding semistandard tableaux. The map is also clearly invertible, so this shows that the map is a bijection.

9. A multiline queue formula for z-deformed probabilites

Recall from Theorem 8.4 that when each yi 0, there is a combinatorial formula for = the steady state probabilities ψw in terms of multiline queues. It is an open problem to extend this result to the case that the yi’s are general. In this section we continue to work in the case yi 0, but we give a generalization of Theorem 8.4 that works for the z-deformed steady= state probabilities. As in the previous section, we allow our states w to be compositions, not just permutations. SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 33

Definition 9.1. For a multiline queue Q, let Qi,r denote the position in the ith column from the right and the rth row from above. We assign a value v Qi,r as follows: ( ) 0 if Q has a ball ⎧ i,r ⎪ v Qi,r ⎪xj if Qi,r is j-covered vacancy ( ) = ⎨ ⎪x otherwise. ⎪ r ⎩⎪ We define the z-weight wtz Q of Q by ( ) wtz Q wt Q zi − v Qi,r , ( ) = ( ) QMi,r Q( ( )) ∈ where wt Q is given by Definition 8.2. And we let ( ) Fw z wtz Q ( ) = Q MLQQ w ( ) ∈ ( ) be the z-weight generating function for multiline queues of type w. Example 9.2. The multiline queue Q in Figure 5 has z-weight 2 2 wtz Q wt Q z1 − x1 z1 − x2 z1 z2 − x1 z2 − x2 z2 z3 − x1 z3 z4 − x1 z4 ( ) = ( )( 2 )( ) ( )( 2 ) ( ) ( ) z5 − x1 z5 z6 − x1 z6 − x3 z6 z7 − x1 z z8 − x1 z8 − x2 z8. ( ) ( )( ) ( ) 7 ( )( ) Theorem 9.3. Let w be a composition, and consider the inhomogeneous TASEP (with yi 0 for all i) on the set of states obtained by permuting the entries of w. Then the= (unnormalized) z-deformed steady state probability of state w is given by

ψw z wtz Q . ( ) = Q MLQQ w ( ) ∈ ( )

Clearly the leading coeﬃcient of wtz Q in the z-variables gives wt Q . So Theorem 9.3 directly implies Theorem 8.4. ( ) ( ) To prove Theorem 9.3, it is enough to show that Fw z satisﬁes the following ( ) exchange equations (see [Can16, Equation (23)]), where si acts on the z-variables. − − zi zi+1 xwi+1 1 si Fw z (9.1) F z if w w + siw ( ) ( −) ( ) i i 1 ( ) = xwi+1 zi zi+1 > (9.2) 1 − si Fw z 0 if wi wi+1. ( ) ( ) = = Remark 9.4. Equations (9.1) and (9.2) are obtained by expanding Equation (23) in [Can16], and (9.1) is a specialization of Proposition 4.2 when yi 0. We need (9.2) to deal with the case that there is more than one particle of a given= type. (In [Can16], Cantini studied the case where there is at most one particle for each type so he included (9.1) as [Can16, Equation (27)] but did not mention (9.2).)

Remark 9.5. One can easily verify that if (9.1) holds, then Fw z + Fs w z is sym- ( ) i ( ) metric in zi and zi+1, i.e. 1 − si Fw z + Fs w z 0. ( )( ( ) i ( )) = Given a composition w, let max w denote its largest part. Note that Fw z is a weighted sum over multiline queues( ) with max w − 1 rows. We will prove( ) Theorem 9.3 by induction on the number of rows;( our( strategy) ) is similar to the one used in [CMW18, Section 3]. For the inductive step, we will view a multiline queue Q with L rows as a multiline queue Q′ with L − 1 rows (the restriction of Q to rows 2 through L) glued on top of a (generalized) multiline queue Q0 with 2 rows (the restriction of Q to rows 1 and 2). 34 DONGHYUNKIMANDLAURENK.WILLIAMS

Definition 9.6. For compositions u and w, we let MLQ u,w denote the set of multiline queues Q of type w whose second row from the bottom( has) type u. We also define u wtz Q Fw ( ′) , = wtz Q ( ) where Q MLQ u,w and Q′ is the multiline queue obtained by deleting the bot- ∈ ( ) u tom row of Q. It is easy to check that Fw does not depend on the choice of Q in MLQ u,w . If MLQ u,w is empty then we set F u 0. ( ) ( ) w = The following lemma follows directly from the definitions.

Lemma 9.7. Consider compositions u u1,...,un and w w1,...,wn such that MLQ u,w is not empty. Pick Q MLQ= ( u,w and) assume= that( Q has L) rows and the number( ) of vacancies in the last∈ row is (t. Then) we have − u t zi v Qi,L Fw x1x2 ...xL zi ( ). = ( ) i ∶ Q Moccupied i ∶ Q Mvacant v Qi,L i,L i,L ( ) Example 9.8. The multiline queue Q in Figure 5 is in MLQ u,w for u 3, 2, 2, 1, 3, 2, 3 and w 2, 2, 1, 4, 4, 4, 2, 3 . By Lemma 9.7, we have ( ) = ( ) = ( ) − − − u 3 z4 x1 z5 x1 z6 x3 Fw x1x2x3 z1z2z3 ( ) ( ) ( )z7z8. = ( ) x1 x1 x3

Remark 9.9. By construction, for ﬁxed i, we can write Fw z as ( ) u u + siu + u (9.3) Fw z FwFu z FwFu z Fw Fsiu z FwFu z . ( ) = Qu ( ) = u,uQi ui 1( ( ) ( )) u,uQi ui 1 ( ) > + = + u siu u siu Our next goal is to analyze the quantities Fw, Fw , Fsiw, Fsiw, case by case.

Lemma 9.10. Let u u1,...,un and w w1,...,wn be compositions. Then we have the following (where= ( indices i)of columns= ( are considered) modulo n).

(1) If there exists i such that ui wi max w then MLQ u,w ∅. < < ( ) ( ) = (2) If there exists i such that ui+1 wi wi+1 max w then MLQ u,w ∅. < < = ( ) ( ) = (3) If there exists i such that wi wi+1 max w and ui wi, MLQ u,w ∅. < < ( ) ≠ ( ) = Proof. The ﬁrst statement above says that in a multiline queue there is no ball which is directly above another ball with a larger label. This statement and the other two follow directly from the bully path algorithm. The following lemma also follows directly from the bully path algorithm.

Lemma 9.11. Let u u1,...,un and w w1,...,wn be compositions. Then we have the following (where= ( indices i)of columns= ( are considered) modulo n).

(1) If ui ui+1, wi wi+1 and ui+1 wi+1 then MLQ u,w ∅ if and only if ≥ ≥ = ( ) ≠ MLQ siu,siw ∅. When they are both nonempty, the matchings of balls between( the bottom) ≠ two rows are preserved. (2) If wi+1 max w or wi+1 min ui,ui+1 then MLQ u,w ∅ if and only = ( ) < ( ) ( ) ≠ if MLQ siu,w ∅. When they are both nonempty, the matchings of balls between( the bottom) ≠ two rows are preserved. (3) If wi max w and wi+1 min ui,ui+1 then MLQ u,w ∅ if and only = ( ) < ( ) ( ) ≠ if MLQ u,siw ∅. When they are both nonempty, the matchings of balls between( the bottom) ≠ two rows are preserved. SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 35

Lemma 9.12. Let u u1,...,un and w w1,...,wn be compositions such that = ( ) = ( ) ui ui+1 and wi wi+1. Then we have the following. > > u siu u siu (1) If wi max w and ui+1 wi+1 then Fw Fw Fsiw Fsiw 0. < ( ) < u = u = siu = = (2) If wi max w and ui+1 wi+1 then F F Fs w 0, = ( ) < w = siw = i = siu ′ ⋅ − and Fw F zi xui+1 zi+1 ′ = ( ) for some F that does not depend on zi and zi+1. u siu (3) If wi max w and ui+1 wi+1 then we have F Fw 0 and we can write < ( ) = siw = = u siu ′ Fw Fsiw F zizi+1 ′ = = for some F that does not depend on zi and zi+1. u siu (4) If wi max w and ui+1 wi+1 then we have F Fs w 0 and we can write < ( ) > siw = i = u siu ′ Fw Fw F zizi+1 ′ = = for some F that does not depend on zi and zi+1. (5) If wi max w and ui+1 wi+1 then we can write = ( ) ′= ′ z − x z + F z + − x z F u i c i 1 siu i 1 c i Fw ( ) , Fsiw ( ) = xc = xc ′ zi+1 − xw 1 ziF u i+ siu ′′ ⋅ − Fsiw ( ) , Fw F zi xwi+1 zi+1. = xwi+1 = ( ) ′ ′′ for some integer c and quantities F , F that do not depend on zi or zi+1. (6) If wi max w and ui+1 wi+1 then we can write = ( ) > ′ ′ z − x z + F z + − x z F u siu i c i 1 u siu i 1 wi+1 i Fw Fw ( ) and Fsiw Fsiw ( ) = = xc = = xwi+1 ′ for some integer c and some F that does not depend on zi or zi+1. Proof. Part (1) follows from Lemma 9.10 (1). We denote max w L. u u siu ( ) = 1 (2) We have Fw Fsiw Fsiw 0 by Lemma 9.10 (1) and (2). Then pick Q = = = 1 ∈ MLQ siu,w (if it is nonempty). Then Qi,L is a vacancy and there is a ball with ( ) 1 1 label ui+1 right above it. So a ui+1-bully path traverses Qi,L. Since Qi+1,L is a ball 1 with label wi+1, no bully path with a label smaller than ui+1 traverses Qi,L. Therefore 1 siu v Qi,L xui+1 . The conclusion for Fw follows from Lemma 9.7. ( ) = u siu (3) By Lemma 9.10 (1) we have F Fw 0. And by Lemma 9.11, MLQ u,w siw = = ( ) is nonempty if and only if MLQ siu,siw nonempty. When they are both nonempty (if they are both empty then we( can simply) take F ′ 0) take Q1 MLQ u,w 2 1 1 2= 2 ∈ ( ) and Q MLQ siu,siw . Then positions Qi,L, Qi+1,L, Qi,L, Qi+1,L are all occupied by ∈ ( ) u siu ′ ′ balls, so by Lemma 9.7, we can write F Fs w F zizi+1 for some F that does not w = i = depend on zi and zi+1. (4) The proof is similar to part (3) but uses Lemma 9.10 (3) and Lemma 9.11 (2). siu (5) The conclusion for Fw follows similarly as in part (2). The other three cases are straightforward using Lemma 9.11 (1), (2), and Lemma 9.7. (6) The proof of this part is straightforward and similar to the previous proofs. The next two lemmas are very similar to Lemma 9.12, but with slightly diﬀerent hypotheses. The proofs are similar so we omit them.

Lemma 9.13. Let u u1,...,un and w w1,...,wn be compositions such that = ( ) = ( ) ui ui+1 and wi wi+1. Then we have the following. = > 36 DONGHYUNKIMANDLAURENK.WILLIAMS

u u (1) If wi max w and ui+1 wi+1 then Fw Fsiw 0. < ( ) ≤ u = u = (2) If wi max w and ui+1 wi+1 then Fw Fsiw 0. = ( ) < u = = (3) If wi max w and ui+1 wi+1 then F 0 and we can write < ( ) > siw = u ′ F F zizi+1 w = ′ for some F that does not depend on zi and zi+1 (4) If wi max w and ui+1 wi+1 then we can write = ( ) ≥ − ′ − ′ u zi xc zi+1F u zi+1 xwi+1 ziF Fw ( ) and Fsiw ( ) = xc = xwi+1 ′ for some integer c and quantity F that does not depend on zi or zi+1.

Lemma 9.14. Let u u1,...,un and w w1,...,wn be compositions such that = ( ) = ( ) ui ui+1 and wi wi+1. Then we have the following. > = u siu (1) If wi max w then F Fw and both are symmetric in zi and zi+1. < ( ) w = (2) If wi max w then either we have = ( ) ′ z − x z + − x F u siu i c i 1 c Fw Fw ( )( 2 ) = = xc ′ for some integer c and quantity F that does not depend on zi or zi+1, OR ′ ′ z − x z + − x F z − x z + − x F u i ui+1 i 1 ui+1 siu i ui+1 i 1 c Fw ( )(2 ) and Fw ( )( ) = xui+1 = xui+1 xc ′ for some integer c and quantity F that does not depend on zi or zi+1. The following lemma is obvious.

Lemma 9.15. Let u u1,...,un and w w1,...,wn be compositions such that = ( u ) = ( ) ui ui+1 and wi wi+1. Then F is symmetric in variables zi and zi+1. = = w Proposition 9.16 will be used together with (9.3) to prove Theorem 9.3.

Proposition 9.16. Let u u1,...,un and w w1,...,wn be compositions, and assume that (9.1) or (9.2) =holds( for u (as) appropriate).= ( Then) we have the following. If wi wi+1 and ui ui+1, we have > > xw 1 zi − zi+1 − u + siu i+ u + siu (9.4) 1 si Fw Fu z Fw Fsiu z ( ) FsiwFu z FsiwFsiu z ( )( ( ) ( )) = zi zi+1 − xw 1 ( ( ) ( )) ( i+ ) If wi wi+1 and ui ui+1, we have > = x z − z + − u wi+1 i i 1 u (9.5) 1 si FwFu z ( ) FsiwFu z . ( ) ( ) = zi zi+1 − xw 1 ( ( )) ( i+ ) If wi wi+1 and ui ui+1, we have = > u siu (9.6) 1 − si F Fu z + F Fs u z 0. ( )( w ( ) w i ( )) = If wi wi+1 and ui ui+1, we have = = u (9.7) 1 − si F Fu z 0. ( ) w ( ) = Proof. We may assume that max w max u + 1 otherwise there is nothing to u siu u siu ( ) = ( ) prove as Fw Fw Fsiw Fsiw 0. We will prove (9.4) using Lemma 9.12. The proofs of the= other three= equations= = are similar using Lemma 9.13, Lemma 9.14 and Lemma 9.15 respectively so we omit them. We divide the proof of (9.4) into six possible cases. SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 37

-Case 1: Suppose wi max w and ui+1 wi+1. Then by Lemma 9.12 (1), we have u u siu siu < ( ) < F F Fw Fs w 0 so both sides of (9.4) are zero. w = siw = = i = -Case 2: Suppose wi max w and ui+1 wi+1. By Lemma 9.12 (2), the right-hand side of (9.4) is zero. Using= Lemma( ) 9.12 (2)< and (9.1), the left-hand side of (9.4) is − − s u ′ zi zi+1 xui+1 1 si Fu z (9.8) 1 − s F i F z 1 − s F z − x z + i w siu i i ui+1 i 1 ( ) ( −) ( ) ( )( ( )) = ( ) ( ) xui+1 zi zi+1 − − − ′ zi xui+1 zi+1 xui+1 1 si Fu z 1 − s F z z + , i i i 1 ( )( ) ( −) ( ) = ( ) xui+1 zi zi+1 which is zero since we are applying 1 − si to a quantity symmetric in zi, zi+1. ( ) -Case 3: Suppose wi max w and ui+1 wi+1. By Lemma 9.12 (3), (9.4) becomes < ( ) = − − ′ ′ zi zi+1 xwi+1 1 si F zizi+1Fu z F z z + F z , or equivalently i i 1 siu ( ) ( )( − ( )) ( ) = xwi+1 zi zi+1 z z + − x 1 − s F z F z i i 1 wi+1 i u , siu ( ) ( −) ( ) ( ) = xwi+1 zi zi+1 which is (9.1) for u and hence true by hypothesis. -Case 4: Suppose wi max w and ui+1 wi+1. The left-hand side of (9.4) is < ( ) > u siu ′ (9.9) 1 − si F Fu z + F Fs u z F zizi+1 1 − si Fu z + Fs u z . ( )( w ( ) w i ( )) = ( )( ( ) i ( )) − + By Remark 9.5, 1 si Fu z Fsiu z 0, so the left-hand side is zero. And by Lemma 9.12 (4),( the right-hand)( ( ) side of( )) (9.4)= is also zero. -Case 5: Suppose wi max w and ui+1 wi+1. Then as shown in (9.8), we have − siu = ( ) = 1 si Fw Fsiu z 0. Therefore, multiplying the left-hand side of (9.4) by a (constant,)( and using( )) Lemma= 9.12 (5), and (9.1), we get − − u + siu − − u zi zi+1 xwi+1 1 si FwFu z Fw Fsiu z zi zi+1 xwi+1 1 si Fw Fu z ( ) ( )( (−) ( )) ( ) ( )(− ( )) xwi+1 zi zi+1 = xwi+1 zi zi+1 ′ ′ zi−xc zi+1F zi+1−xc ziF z z + − x ( ) Fu z − ( ) siFu z i i 1 wi+1 ( xc ( ) xc ( )) ( ) − = xwi+1 zi zi+1 − ′ − ′ zi+1 xc ziF zi+1 xc ziF ′ z z + − x ( ) Fu z − ( ) siFu z z + − x z F i i 1 wi+1 ( xc ( ) xc ( )) + i 1 wi+1 i F z ( ) − ( ) u = xwi+1 zi zi+1 xwi+1 ( ) ′ ′ z + − x z F z z + − x 1 − s F z z + − x z F i 1 c i i i 1 wi+1 i u + i 1 wi+1 i F z ( ) ( ) ( −) ( ) ( ) u = xc xwi+1 zi zi+1 xwi+1 ( ) ′ ′ zi+1 − xc ziF zi+1 − xw 1 ziF + i+ u + siu ( ) Fsiu z ( ) Fu z FsiwFu z FsiwFsiu z . = xc ( ) xwi+1 ( ) = ( ) ( )

-Case 6: Suppose wi max w and ui+1 wi+1. Then by Lemma 9.12 (6), we have = ( ) > − − u + siu zi zi+1 xwi+1 1 si FwFu z Fw Fsiu z ( ) ( )( (−) ( )) xwi+1 zi zi+1 − ′ − zi xc zi+1F + z z + − x 1 si ( ) Fu z Fsiu z i i 1 wi+1 ( )( xc ( ( ) ( ))) ( ) − = xwi+1 zi zi+1

zi zi+1 − xw 1 i+ ′ + u + siu ( )F Fu z Fsiu z FsiwFu z FsiwFsiu z , = xwi+1 ( ( ) ( )) = ( ) ( ) + where we used the fact that Fu z Fsiu z is symmetric in zi and zi+1 (Remark 9.5). ( ( ) ( )) 38 DONGHYUNKIMANDLAURENK.WILLIAMS

Now we are ready to prove Theorem 9.3.

Proof of Theorem 9.3. For a composition w w1,...,wn , it is enough to show that w satisﬁes (9.1) or (9.2). We use induction on= (max w . ) Consider the base case max w 2. There is only( one) multiline queue (consisting of one row) of type w, so we have( ) =

Fw z zi zi − x1 . ( ) = i,wMi 1( ) i,wMi 2( ) = = ′ If wi wi+1, (9.2) is immediate. If wi wi+1, we can write Fw z F ⋅ zi −x1 zi+1 and = ′ − ′ > ( ) = ( ) Fsiw z F zi zi+1 x1 for some F that does not depend on zi and zi+1. Therefore the right-hand( ) = ( side of (9.1)) becomes ′ zi zi+1 − x1 1 − si F zi − x1 zi+1 zi zi+1 − x1 ′ ′ ( ) ( )( ( ) ) ( )F x1 F zi zi+1 − x1 x1 zi − zi+1 = x1 = ( ) which is Fsiw z . Now consider( ) the case max w 2 and assume that (9.1) and (9.2) are true for u ( ) > such that max u max w − 1. If wi wi+1 then (9.1) follows from (9.3), (9.4) and ( ) = ( ) > (9.5). If wi wi+1 then (9.2) follows from (9.3), (9.6) and (9.7). = 10. The proof of Theorem 1.3, the monomial factor conjecture In this section we will prove Theorem 1.3, which gives a formula for the largest monomial in x that divides ψw in the setting where yi 0. We will first give an algebraic argument using the isobaric divided difference operators= to show that η w n−2 αi w +⋯+αn−2 w ( ) = i 1 xi ( ) ( ) always divides ψw, see Proposition 10.9. We will then use our combinatorial∏ = formula for z-deformed steady state probabilities in terms of multiline queues (Theorem 9.3) to show that no greater monomial in x divides ψw. Definition 10.1. For a (multivariable) polynomial p and a variable x, we write xd p if xd p and xd+1 ∤ p, that is, d is the highest power of x that divides p. SS S Recall from Definition 1.2 that for w w1,...,wn St n , αi w is the number = ( ) ∈ ( ) ( ) of integers greater than i + 1 among i + 1 wr,wr+1,...,ws−1,ws i , where the subscripts are taken modulo n. To prove{ Theorem= 1.3, we will show= that}

αi w +⋯+αn−2 w x ( ) ( ) ψw, for 1 i n − 2. i SS ≤ ≤ Definition 10.2. Given an integer vector b1, ⋯, bn−2 such that 0 bi i, we define 0 ( ) ≤ ≤ S b1, ⋯, bn−2 St n as follows. Let w 1, 2,...,n St n . Then for 1 i n(− 2, we recursively) ∈ ( ) construct from wi−1=a( state wi )whose∈ ( first) n − i − 1 letters≤ ≤ i−1 are 1, 2,...,n − i − 1, by taking the last bi letters of w and inserting them after 1, 2,...,n − i − 1. That is, i i−1 i−1 i−1 i−1 i−1 w 1, 2,...,n − i − 1,w − + ,...,w − ,w , n − i, w − + ,...,w − . = ( n bi 1 n 1 n n i 1 n bi ) n−2 We set S b1, ⋯, bn−2 w . ( ) = Example 10.3. Using Definition 10.2, we obtain S 1, 1, 2, 3 w4 St 6 as follows. ( ) = ∈ ( ) w0 1, 2, 3, 4, 5, 6 w1 1, 2, 3, 4, 6, 5 w2 1, 2, 3, 5, 4, 6 w3 = (1, 2, 4, 6, 3, 5) w4 = (1, 6, 3, 5, 2, 4) = ( ) = ( ) = ( ) i i−1 Remark 10.4. It follows from Definition 10.2 that αn−1−i w αn−1−i w −bi and i−1 i ( ) = ( ) αj w αj w if j n − 1 − i. ( ) = ( ) ≠ SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 39

Lemma 10.5. Let w S b1, ⋯, bn−2 St n . Then we have = ( ) ∈ ( ) αi w n − 1 − i − bn−1−i. ( ) = n−2 And for any w St n , there exists b1, ⋯, bn−2 Z such that 0 bi i and ∈ ( ) ( ) ∈ ≤ ≤ w S b1, ⋯, bn−2 . ∼ ( ) 0 Proof. Since αi w n − 1 − i, Remark 10.4 implies that αi w n − 1 − i − bn−1−i. ( ) = ( ) = There are a total of n − 1 ! possible states S b1, ⋯, bn−2 , which are all cyclically ( ) ( ) diﬀerent since the αi w ’s are cyclically invariant. So they cover every element in St n up to a cyclic equivalence.( ) ( ) Example 10.6. Continuing Example 10.3, if w ∶ S 1, 1, 2, 3 1, 6, 3, 5, 2, 4 then = ( ) = ( ) α1 w 1, α2 w 1, α3 w 1 and α4 w 0, as predicted by Lemma 10.5. ( ) = ( ) = ( ) = ( ) = Lemma 10.7. If ψw′ z πk a, b ψw z , and if a monomial M in x1,...,xn−1 divides M ( ) = ( ) ( ) ψw z , then divides ψw′ z . ( ) xb ( ) zk zk+1−xb f z −skf z Proof. This follows from the fact that πk a, b f z ( ) ( )− ( ) , and each ( ) ( ) = xb zk zk+1 ψw z is a polynomial (Proposition 4.2). ( ) ′ ′ Lemma 10.8. Consider w,w St n such that w 1,...,i,wi+1,...,wn and w ∈ ( ) = (′′ ) = 2,...,i,wn,wi+1,...,wn−1, 1 for 1 i n − 1. Let w 1,...i,wn,wi+1,...,wn−1 . ( ) ≤ ≤ M = ( ) If a monomial M in x1,...,xn−1 divides ψw z , divides both ψw′ z and ψw′′ z . ( ) x1⋯xi ( ) ( ) Proof. Using Lemma 10.7 and the fact that

ψw′ z πi−1 wn, i ⋯π2 wn, 3 π1 wn, 2 πn wn, 1 ψw z , ( M) = ( ) ( ) ( ) (M ) ( ) we conclude that divides ψ ′ z . The fact that divides ψ ′′ z follows x1⋯xi w x1⋯xi w from Proposition 4.2 since w′ and w(′′ )are cyclically equivalent. ( ) Proposition 10.9. For w St n , we have ∈ ( ) αi w +⋯+αn−2 w x ( ) ( ) ψw, for 1 i n − 2. i S ≤ ≤ Proof. By Proposition 4.2, it is enough to show this for any cyclic shift of w, so we may assume w S b1, ⋯, bn−2 for some vector b1, ⋯, bn−2 . Using the notation in Deﬁnition 10.2,= we( will use induction) on i to show( that ) i i i α1 w α2 w αn−2 w (10.1) x ( ) x1x2 ( )⋯ x1⋯xn−2 ( ) ψwi z 1 ( ) ( ) S ( ) 0 Since w is the identity permutation, we know from the expression for ψ 1,2,...,n z n−2 n−3 1 ( )( ) in Proposition 4.2 that x1 x1x2 ⋯ x1⋯xn−2 divides ψw0 z . Now assume (10.1) holds for i − 1. By repeatedly( applying) ( Lemma) 10.8, we have( that) 1 α1 wi− i−1 i−1 α2 w αn−2 w x1 ( ) x1x2 ( )⋯ x1⋯xn−2 ( ) ( ) ( b ) x1⋯xn−1−i i ( ) divides ψwi z . Now Remark 10.4 implies that 1 ( ) α1 wi− i−1 i−1 ( ) α2 w ⋯ ⋯ αn−2 w i x1 x1x2 ( ) x1 xn−2 ( ) α1 w α2 wi α 2 wi ( ) ⋯ ⋯ − n− ( ) ( b ) x1 x1x2 ( ) x1 xn 2 ( ). x1⋯xn−1−i i = ( ) ( ) ( ) Thus we conclude (10.1) is true for i. And (10.1) for i n − 2 gives the claim. = Having proved Proposition 10.9, our next goal is to show that no larger monomial in x divides ψw. d d Lemma 10.10. If x ψw then x ψw z . i SS i SS ( ) 40 DONGHYUNKIMANDLAURENK.WILLIAMS

d d Proof. If xi ψw z then xi also divides ψw LCz ψw z . For the converse, assume d S ( ) d = ( ( )) xi ψw. Then by Theorem 8.4, xi wt Q for every multiline queue Q MLQ w . S d S ( ) ∈ ( ) Since wt Q wtz Q we have xi wtz Q for every Q MLQ w , which implies d ( ) S ( ) S ( ) ∈ ( ) x ψw z by Theorem 9.3. i S ( ) ˆ Lemma 10.11. Given a multiline queue Q MLQ w1, ⋯,wn, 1 , there exists Q d ∈ ( )d ˆ ∈ MLQ w1 + 1, ⋯,wn + 1, 2, 1 such that x wt Q if and only if x wt Q . ( ) 1 SS ( ) 1 SS ( ) Proof. Recall the notion of the type of a row of a multiline queue from Definition 8.1. ˆ We will construct Q so that if the rth row of Q has type u1, ⋯,un, 1 (up to a cyclic ˆ ( ) shift), the r + 1 st row of Q has type u1 + 1, ⋯,un + 1, 2, 1 (up to a cyclic shift). ( ) ( ) Let Qˆ0 be the two-row multiline queue with n + 2 columns, whose first and second rows have types 2, ⋯, 2, 1, 2, 2 and 3, ⋯, 3, 2, 1 . For 1 i n − 1, we inductively ( ) ( ) ≤ ≤ construct Qî as follows: ● Cyclically shift the columns of Q so that the leftmost entry in the ith row of the ′ resulting multiline queue Q is a ball labeled 1. Let v1,...,vk, 1, vk+1,...,vn be the type of the i + 1 st row of Q′. Add a new( i + 2 nd row of type) ( ) î−1 ( ) v1 +1, ⋯, vk +1, 2, 1, vk+1 +1, ⋯, vn +1 to Q , then cyclically shift the columns (of the multiline queue so that the leftmost) entry in the i + 2 nd row is a ball ( ) labeled 1. Denote the resulting multiline queue by Qî. We set Qˆ Qˆn−1. =

1 1 2 1 1 2 1 2 3

Q Qˆ0

1 1 1 2 1 2 2 3 1 1 2 3 1 2 3 4

Qˆ1 Qˆ2 Qˆ = Figure 9. An example of the construction in Lemma 10.11 with n 3. =

By Deﬁnition 8.2, the exponent of x1 in wt Q is the number of vacancies in rows 2,...,n minus the number of 1-covered vacancies( ) in rows 2,...,n. By construction, the number of vacancies and 1-covered vacancies in the rth row of Q agrees with the corresponding number of vacancies in the r + 1 st row of Qˆ for r 2. All the ( ) ≥ vacancies in row 2 of Qˆ are 1-covered, so the exponents of x1 in wt Q and wt Qˆ are the same. ( ) ( ) SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 41

Example 10.12. Figure 9 shows a multiline queue Q of type 4, 3, 2, 1 as well as ( ) Qˆ0, Qˆ1, and Qˆ2 Qˆ, as constructed in the proof of Lemma 10.11. We have = 1+2 1 x2 2 x3 3 wt Q x1 x2 x2x3 ( ) = ( ) (x1 ) (x1 ) = ˆ 1+2+3 1+2 1 x2 3 x3 2 x4 6 3 wt Q x1 x2 x3 x2x3x4. ( ) = ( ) (x1 ) (x1 ) (x1 ) =

Note that the exponents of x1 in wt Q and wt Qˆ are both zero. ( ) ( ) Proposition 10.13. For the state w St n , we have ∈ ( ) α1 w +⋯+αn−2 w x ( ) ( ) ψw. 1 SS α1 w +⋯+αn−2 w Proof. By Proposition 10.9, we know that x ( ) ( ) ψw. So we need to show 1 S that no greater power of x1 divides ψw. We use induction, and suppose this is true for states in St n . For the sake of contradiction, suppose there exists a state ( ) α1 w +⋯+αn−1 w +1 α1 w +⋯+αn−1 w +1 w St n + 1 such that x ( ) ( ) ψw. By Lemma 10.10, x ( ) ( ) ∈ ( ) 1 S 1 S ψw z . Cyclically shifting the state if necessary, we can assume the last component ( ) 5 of w is a 1; so write w w1, ⋯,wn−k−1, 2,wn−k, ⋯,wn−1, 1 . Then we have = ( ) ⋯ ψ w1,⋯,w 1,2,1,w ,⋯,wn−1 z πn−k+1 wn−k, 1 πn wn−1, 1 ψw z , ( n−k− n−k )( ) = ( ) ( ) ( ) α1 w +⋯+αn−1 w +1−k ( ) ( ) which implies that x1 ψ w1,⋯,wn−k−1,2,1,wn−k,⋯,wn−1 z (by Lemma 10.7). ′ S ( )( ) Applying a cyclic shift, we let w wn−k, ⋯,wn−1,w1, ⋯,wn−k−1, 2, 1 ; we also have α1 w +⋯+αn−1 w +1−k =′ ( ′ ) ′ x ( ) ( ) ψw′ . Since α1 w α1 w −k and α2 w α2 w , ⋯,αn−1 w 1 S ( ) = ( ) ( ) = ( ) ( ) = αn−1 w , we have ( ) ′ ′ a1 w +⋯+an−1 w +1 ′ (10.2) x ( ) ( ) ψw′ for w wn−k, ⋯,wn−1,w1, ⋯,wn−k−1, 2, 1 St n + 1 . 1 S = ( ) ∈ ( ) ′′ Consider the state w wn−k − 1, ⋯,wn−1 − 1,w1 − 1, ⋯,wn−k−1 − 1, 1 St n . By ′′ ′′ = (α1 w +⋯+αn−2 w ) ∈ ( ) the induction hypothesis, x1 ( ) ( ) ψw′′ , so there exists a multiline queue Q of type w′′ such that SS ′′ ′′ α1 w +⋯+αn−2 w x ( ) ( ) wt Q . 1 SS ( ) By Lemma 10.11, there is a multiline queue Q′ of type w′ such that

′′ ′′ α1 w +⋯+αn−2 w ′ x ( ) ( ) wt Q . 1 SS ( ) ′ ′ ′′ ′ ′′ Since α1 w 0 and α2 w α1 w , ⋯,αn−1 w αn−2 w , we have ( ) = ( ) = ( ) ( ) = ( ) ′ ′ α1 w +⋯+αn−1 w ′ x ( ) ( ) wt Q . 1 SS ( ) This contradicts (10.2). Lemma 10.14. For w St n and 1 i n − 3, let w′ St n − i be obtained by erasing 1, ⋯, i in w and∈ decreasing( ) the≤ other≤ integers by i∈. Given( )Q′ MLQ w′ , d ′ d ∈ ( ) there exists Q MLQ w such that x wt Q if and only if x + wt Q . ∈ ( ) 1 SS ( ) i 1 SS ( ) Proof. We increase the label of each ball in Q′ by i, and insert trivial bully paths of type 1,...,i in the appropriate positions to get a multiline queue of type w.

5 It is possible that w = (w1,...,wn 1, 2, 1), in which case the following argument simpliﬁes: there − are no π operators in the following line, and we take w′ = w. 42 DONGHYUNKIMANDLAURENK.WILLIAMS

Example 10.15. Let w 1, 4, 2, 7, 6, 5, 3 St 7 and i 3. Then w′ 1, 4, 3, 2 . Figure 10 shows Q′ MLQ= (w′ on the left) and∈ (Q)on the right.= The weights= ( are ) ∈ ( ) ′ 1+2 1 x2 2 x3 3 wt Q x1 x2 x2x3 ( ) = (x1 ) (x1 ) = 1+2+3+4+5 1+2+3+4 1+2+3 1+2 1 x5 2 x6 15 10 6 3 wt Q x1 x2 x3 x4 x5 x1 x2 x3x5x6. ( ) = (x4 ) (x4 ) = ′ The exponent of x1 in wt Q and the exponent of x4 in wt Q are both zero. ( ) ( ) 1 2 1 3 2 1 1 3 4 2 1 1 2 3 4 2 5 1 2 3 1 3 5 6 2 4 1

Q′ Q

Figure 10. An example of Lemma 10.14, with w 1, 4, 2, 7, 6, 5, 3 , i 3, and w′ 1, 4, 3, 2 . = ( ) = = ( ) Proof of Theorem 1.3. For w St n and a number 1 i n − 3, consider the state w′ St n − i obtained by erasing∈ ( )1, ⋯, i in w and decreasing≤ ≤ other integers by ′ ′ ∈ ( ) α1 w +⋯+αn−i−2 w i. By Proposition 10.13, x1 ( ) ( ) ψw′ , hence by Theorem 8.4, there exists a multiline queue Q′ of type w′ such that SS

′ ′ α1 w +⋯+αn i 2 w ′ x ( ) − − ( ) wt Q . 1 SS ( ) By Lemma 10.14, there exists Q MLQ w such that ∈ ( ) ′ ′ α1 w +⋯+αn−i−2 w (10.3) x + ( ) ( ) wt Q . i 1 SS ( ) By the construction of w′ we have

′ ′ ′ αi+1 w α1 w , αi+2 w α2 w , ... αn−2 w αn−i−2 w , ( ) = ( ) ( ) = ( ) ( ) = ( ) so (10.3) implies

αi+1 w +⋯+αn−2 w x + ( ) ( ) wt Q . i 1 SS ( ) By Proposition 10.9 we have that

αi+1 w +⋯+αn−2 w x + ( ) ( ) ψw i 1 SS for 1 i n − 3. Combining this with Proposition 10.13, we conclude that the largest ≤ ≤ n−2 αi w +⋯+αn−2 w monomial that can be factored out of ψw is xi ( ) ( ), as desired. Recall i 1 ′ ∏= ′ from Lemma 10.5 that if two states w,w St n have the same αi w αi w for all 1 i n − 2 then w w′. This completes∈ the( proof.) ( ) = ( ) ≤ ≤ ∼ SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 43

11. Future questions There are various natural questions that arise from this work. Problem 11.1. Find a combinatorial formula for steady state probabilities of the inhomogeneous TASEP (e.g. using multiline queues) in the case where yi 0. ≠ Problem 11.2. Find a purely combinatorial proof of Theorem 1.3.

Problem 11.3. Prove that for any permutation w, the steady state probability ψw in the inhomogeneous TASEP can be written as a positive sum of double Schubert polynomials Sw x1,...,xn; y1,...,yn multiplied by some linear factors xi − yj . ( ) ( ) Problem 11.3 is open even when yi 0 for all i. One way to approach this problem would be to ﬁnd a geometric interpretation= of each steady state probability. Recall that Schubert polynomials represent cohomology classes of Schubert varieties in the complete ﬂag variety. Can one associate a variety to each state whose cohomology class is represented by the corresponding (unnormalized) steady state probability? 12. Appendix: technical results for the proof of Theorem 5.9 In this section, we collect some technical results (Proposition 12.4 and Proposition 12.5) which we will use for the proof of Theorem 5.9.

Lemma 12.1. [Can16, Lemma 22] Let K x1; x2, ⋯, xm+1 be a rational expression in ( ) x1,...,xm+1 which is symmetric in the variables x2, ⋯, xm+1. We have m+1 K x ; x , ⋯, xˆ , ⋯, x + ∂ ⋯∂ K i 1 i m 1 . m 1 ( m+1 ) = Qi 1 = xi − xj j∏1 ( ) j=i ≠ Now we rewrite (5.1). In what follows, we let m denote 1, 2,...,m . [ ] { } Lemma 12.2. For λ Val n , let mul λ k for some k. Let λ˜ denote the partition obtained by deleting the∈ first( )k parts of(λ).≥ Then Sn z; x; y is equal to λ( ) n−k λ1−λ 1+k S σ k+ z ; xˆ; y x − y z − x λ˜ I i l i j ( ( ) ) 1 l n−∏mul(λ)( ) 1 i λ1∏−λk+1+k ( ) ≤ ≤ i I j [n−≤λ≤1−mul(λ)+k]∖I ∈ ∈ − I [n−λ1−Qmul(λ)+k] xi xj ⊆ SIS k i∏I ( ) j [n−λ1−mul∈ (λ)+k]∖I = ∈ Proof. Note that the expression in Lemma 12.2 implies that Sn z; x; y is symmetric λ( ) in variables x1, ⋯, xn−λ1 −mul(λ)+k. Since we can take k mul λ , Lemma 12.2 implies Sn ⋯ = ( ) that λ z; x; y is symmetric in variables x1, , xn−λ1 . Now we use induction on n. Suppose( the statement) holds for partitions in Val n − 1 . Let λ′ to be the partition obtained by deleting the( first) part of λ. By the induction n−1 S ⋯ ′ hypothesis, λ′ z; x; y is symmetric in variables x1, , x(n−1)−λ1 . Note that we have ′ ( ) n − 1 − λ1 n − λ1 − mul λ . Thus we rewrite (5.1) using Lemma 12.1 as follows ((12.1) ) ≥ ( ) n−mul(λ) (λ1−λ2+1) n−λ1−mul(λ)+1 Sn−1 λ1−λ2+1 − − λ′ σ z ; xî; y xi yl zl xj n−λ1−mul(λ)+1 ( ( ) ) l∏1 ( ) l∏1 j∏1 ( ) = = j=i Sn z; x; y ≠ , λ n−λ1−mul(λ)+1 ( ) = Qi 1 = xi − xj j∏1 ( ) j=i ≠ 44 DONGHYUNKIMANDLAURENK.WILLIAMS which implies the k 1 case of Lemma 12.2. Assume k 1. In this= case we have mul λ 1, which implies mul λ′ mul λ − 1 ′ > ( ) n>−1 ( ) = ( ) and λ λ2 λ1. Applying Lemma 12.2 to S ′ z; x; y and k − 1 gives 1 = = λ ( ) ( ) (12.2)

n−1 S ′ z; x; y λ ( ) = n−k λ1−λ 1+k−1 S σ k+ z ; xˆ; y x − y z − x λ˜ I i l i j ( ( ) ) 1 l n−∏mul(λ)( ) 1 i λ1−∏λk+1+k−1 ( ) ≤ ≤ i I j [n−≤1≤−λ1−mul(λ)+k]∖I ∈ ∈ , − I [n−1−λQ1−mul(λ)+k] xi xj ⊆ SIS k−1 i∏I ( ) j [n−1−λ1−∈mul(λ)+k]∖I = ∈ and rewriting (12.1) using the fact λ1 λ2 gives = (12.3) n−mul(λ) n−λ1−mul(λ)+1 Sn−1 − − λ′ σ z ; xˆj ; y xj yl z1 xm n−λ1−mul(λ)+1 ( ( ) ) l∏1 ( ) m∏1 ( ) = m=j Sn z; x; y ≠ . λ n−λ1−mul(λ)+1 ( ) = jQ1 = xj − xm m∏1 ( ) m=j ≠ Plugging (12.2) into (12.3) gives (12.4) n−λ1−mul(λ)+1 n−mul(λ) − + Sn λ1 λk+1 k λ z; x; y fI σ z ; xˆj; y MI z; xˆj ; y xj − yl ( ) = I [n−1−λQ1−mul(λ)+k] jQ1 ( ( ) ) ( ) Ml 1 ( ) = = ⊆ SIS k−1 = where

n−k fI z; x; y S z; xˆ; y ( ) = λ˜ ( I ) n−λ1−mul(λ) xi − yl zi − xm z1 − xm 1 l n−mul(λ)( ) 2 i λ1−λk 1+k ( ) m 1 ( ) ∏ ∏ + ∏= ≤ ≤ i I m [n−≤1−≤λ1−mul(λ)+k]∖I MI z; x; y ∈ ∈ . ( ) = n−λ1−mul(λ) xi − xm xj − xm i∏I ( ) m∏1 ( ) m [n−1−λ1−∈mul(λ)+k]∖I = ∈

For a ﬁxed I0 i1 ⋯ ik n − λ1 − mul λ + k , to have = { < < } ⊆ [ ( ) ] λ1−λk+1+k n−k λ1−λk+1+k fI σ z ; xˆ; y S σ z ; x ˆ ; y ( ( ) j ) = λ˜ ( ( ) I0 ) we need to take I Ih i1 ⋯ ih−1 ih+1 − 1 ⋯ ik − 1 and j ih = = { < < < n−k <λ1−λ <1+k } = ≤ n − λ − mul λ + 1. So taking the coeﬃcient of S σ k+ z ; x ˆ ; y in (12.4) 1 λ˜ I0 gives ( ) ( ( ) ) n−mul(λ) M z; x ; y x − y . Ih iˆh ih l hQ1 ( ) Ml 1 ( ) i n−λ1−≥mul(λ)+1 = h≤ SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 45

Note that we have n−mul(λ) M z; x ; y x − y Ih iˆh ih l ( ) Ml 1 ( ) = n−λ1−mul(λ)+1 xi − yl zi − xm z1 − xm 1 l n−mul(λ)( ) 2 i λ1−λk 1+k ( ) m 1 ( ) ∏ ≤ ≤ ∏ + ∏= ≤ ≤ i I0 m [n−λ1−mul(λ)+k]∖I0 m ih ∈ ∈ ≠ . = n−λ1−mul(λ)+1 − − xi xm xih xm i I0,i ih ( ) m 1 ( ) ∏ ∏= m [n−λ1∈−mul≠(λ)+k]∖I0 m ih ∈ ≠ We claim (12.5) xi − yl zi − xm 1 l n−mul(λ)( ) 1 i λ1−λk 1+k ( ) ∏ ≤ ≤ ∏ + n−mul(λ) ≤ ≤ i I0 m [n−λ1−mul(λ)+k]∖I0 ∈ ∈ M z; x ; y x −y . Ih iˆh ih l xi − xm = hQ1 ( ) Ml 1 ( ) i∏I0 ( ) ih n−λ1−≥mul(λ)+1 = m [n−λ1−mul∈ (λ)+k]∖I0 ≤ ∈ We view both sides as polynomials in z1 of degree at most n − λ1 − mul λ . So it ( ( )) is enough to show that they coincide when we plug in z1 x1, ⋯, xn−λ1−mul(λ)+1. If we = plug in z xm for m I0 then both sides are zero. If we plug in z1 xi for some = ∉ = t it I0 then only h t on the right hand side does not vanish and we have ∈ = xi − yl zi − xm 1 l n−mul(λ)( ) 2 i λ1−λk 1+k ( ) n−mul(λ) ∏ ≤ ≤ ∏ + ≤ ≤ i I0 m [n−λ1−mul(λ)+k]∖I0 M z; x ; y x −y ∈ ∈ . It iˆt it l z1 xit ( ) Ml 1 ( )S = = xi − xm = i I∏0,i it ( ) m [n−λ1∈−mul≠(λ)+k]∖I0 ∈

This is the same as the left hand side of (12.5) evaluated at z1 xit . In conclusion (12.4) becomes = Sn z; x; y λ( ) = n−k λ1−λ 1+k S σ k+ z ; x ˆ ; y x − y z − x λ˜ I0 i l i m ( ( ) ) 1 l n−∏mul(λ)( ) 1 i λ1∏−λk+1+k ( ) ≤ ≤ i I0 m [n−≤λ≤1−mul(λ)+k]∖I0 ∈ ∈ , − I0 [n−λ1Q−mul(λ)+k] xi xm ⊆ i I0 SI0S k ∏ ( ) = m [n−λ1−mul∈ (λ)+k]∖I0 ∈ which completes the proof.

Proposition 12.3. Choose λ Val n and a such that 1 a n − λ1. If we specialize ∈ ( ) n ≤ ≤ z1 ∶ xa in the z-Schubert polynomial S z; x; y , we obtain = λ( ) Sn (12.6) λ z; x; y z1 xa ( ))S = = n−mul(λ) (λ1−λ2+1) n−λ1−mul(λ)+1 Sn−1 λ1−λ2+1 λ′ σ z ; xaˆ; y xa − yl zi − xm , ( ( ) )) Ml 1 ( ) Mi 2 mM1 ( ) = = m=a ≠ where λ′ is the partition obtained by deleting the ﬁrst part of λ. If length λ 1 then ( ) = we regard λ2 0. = Sn Proof. Taking k mul λ in (12.2), we see that λ z; x; y is symmetric in variables = ( ) ( ) x1, ⋯, xn−λ1 . So it is enough to show (12.6) for a 1, which follows from (12.1). = 46 DONGHYUNKIMANDLAURENK.WILLIAMS

Note that in Proposition 12.4 and 12.5, the operator ∂i acts on z-variables.

′ Proposition 12.4. Let λ Val n such that λ1 λ2, and write λ λ1,λ for some λ′. Then ∈ ( ) > = ( )

λ1 −λ2+1 Sn − (12.7) (λ1,λ′) z; x; y zi xn+1−λ1 ( ) Mi 3 ( ) = ′ n−λ1 mul((λ1−1,λ ))−1 Sn − − ∂1 (λ1−1,λ′) σ z ; x; y z1 xi z2 yn−i . = ( ( ) ) Mi 1 ( ) Mi 1 ( ) = = Proof. We will refer to the left and right-hand sides of (12.7) as LHS and RHS. To prove they are equal, we will view each side as a polynomial in z1 and analyze its Sn degree to use interpolation. Note that ′ σ z ; x; y does not depend on z1 so (λ1−1,λ )( ( ) ) exchanging variables z1 and z2 is the same as plugging in z1 in place of z2. So we get

′ 1 n−λ1 mul((λ1−1,λ ))−1 Sn − − RHS (λ1−1,λ′) σ z ; x; y z1 xi z2 yn−i = z1 − z2 ( ( ) ) Mi 1 ( ) Mi 1 ( ) = = ′ n−λ1 mul((λ1−1,λ ))−1 − Sn − − (λ1−1,λ′) σ z ; x; y z2 z1 z2 xi z1 yn−i . ( ( ) )S = Mi 1 ( ) Mi 1 ( ) = = n ′ By (5.1), S z; x; y has degree at most n − λ1 − 1 − mul λ − 1,λ in z1. (λ1−1,λ)( ) ( ( ) (( ))) So the numerator of RHS has degree at most n − λ1 in z1, which implies that ( ) RHS has degree at most n − λ1 − 1 in z1. Meanwhile, LHS has degree at most ( ) n − λ1 − mul λ n − λ1 − 1 in z1 by (5.1). So it is enough to show that LHS and ( ( )) = ( ) RHS coincide when we plug in z1 xh for any 1 h n − λ1. We have = ≤ ≤ ′ n−λ1 mul((λ1−1,λ ))−1 n −S σ z ; x; y z − x x − y − (λ1−1,λ′) z2 xh 2 i h n i ( ( ) )S = i∏1 ( ) i∏1 ( ) RHS z1 xh = = S = = xh − z2 ′ n−λ1 mul((λ1−1,λ ))−1 Sn − − (λ1−1,λ′) σ z ; x; y z2 xh z2 xi xh yn−i . = ( ( ) )S = Mi 1 ( ) Mi 1 ( ) i=h = ≠ And by Proposition 12.3 we have Sn λ1−1,λ′ z; x; y z1 xh ( )( ))S = n−mul((λ1−1,λ′)) (λ1−λ2) n−λ1−mul((λ1−1,λ′))+2 Sn−1 λ1−λ2 − − λ′ σ z ; xhˆ ; y xh yl zi xm = ( ( ) )) Ml 1 ( ) Mi 2 mM1 ( ) = = m=h ≠ ′ n−mul((λ1−1,λ )) (λ1−λ2) n−λ1+1 Sn−1 λ1−λ2 − − λ′ σ z ; xhˆ ; y xh yl zi xm , = ( ( ) )) Ml 1 ( ) Mi 2 mM1 ( ) = = m=h ≠ ′ where the last equality uses the fact that when mul λ1 −1,λ 1 we have λ1 −λ2 1 (( )) > = so the product over i 2 to λ1 − λ2 is vacuous. Shifting variables by z → σ z we deduce = ( )

′ n−mul((λ1−1,λ )) (λ1−λ2+1) n−λ1−1 Sn Sn−1 λ1−λ2+1 − − (λ1−1,λ) σ z ; x; y z2 xh λ′ σ z ; xhˆ ; y xh yl zi xm . ( ( ) )S = = ( ( ) ) Ml 1 ( ) Mi 3 mM1 ( ) = = m=h ≠ SCHUBERT POLYNOMIALS, THE TASEP, AND EVIL-AVOIDING PERMUTATIONS 47

Plugging this into RHS z1 x gives S = h n−1 n−λ1 (λ1−λ2+1) n−λ1+1 Sn−1 λ1−λ2+1 − − − RHS z1 xh λ′ σ z ; xhˆ ; y xh yi z2 xi zi xm S = = ( ( ) ) Mi 1 ( ) Mi 1 ( ) Mi 3 mM1 ( ) = i=h = m=h ≠ ≠ n−1 (λ1−λ2+1) n−λ1 λ1−λ2+1 Sn−1 λ1−λ2+1 − − − λ′ σ z ; xhˆ ; y xh yi zi xm zi xn+1−λ1 = ( ( ) ) Mi 1 ( ) Mi 2 mM1 ( ) Mi 3 ( ) = = m=h = ≠ λ1−λ2+1 Sn − (λ1,λ′) z; x; y z1 xh zi xn+1−λ1 LHS z1 xh . = ( )S = Mi 3 ( ) = S = =

b ˜ Proposition 12.5. Choose λ Val n with mul λ b 1. Write λ λ1 , λ for some λ˜. We have ∈ ( ) ( ) = > = (( ) )

b−1 b+λ1−λ˜1 n (12.8) S z; x; y zi − yn+1−b zi − xn+1−λ1 ((λ1)b,λ˜) ( ) Mi 1( ) iMb+2 ( ) = = mul((λ1−1,λ˜))−1 n ∂b ⎛S 1 z; x; y zb+1 − yn+1−b−i ⎞ ((λ1)b− ,λ1−1,λ˜) = ( ) Mi 1 ( ) ⎝ = ⎠ Proof. We view both sides as polynomials in z1 and analyze their degrees to use inter- n polation. By (5.1), S z; x; y is of degree at most n−λ1 −b in z1 so the right ((λ1)b,λ˜)( ) ( ) Sn hand side is of degree at most n − λ1 − b + 1 in z1. Likewise, 1 z; x; y ( ) ((λ1)b− ,λ1−1,λ˜)( ) is of degree at most n − λ1 − b + 1 in z1 so the right hand side is of degree at most ( ) n − λ1 − b + 1 in z1. Since b 1, it is enough to show that they coincide when we ( ) > plug in z1 xa for 1 a n − λ1. = ≤ ≤ We use induction on b. First assume b 2. Setting z1 ∶ xa on both sides, Proposition 12.3 gives = =

n−1 2+λ1−λ˜1 n−1 LHS z1 x S z; xaˆ; y xa − yl zi − xn+1−λ1 a (λ1,λ˜) S = = ( ) Ml 1 ( ) Mi 4 ( ) = = n−1 n−λ1 mul((λ1−1,λ˜))−1 ⎛ n−1 ⎞ RHS z1 x ∂2 S σ z ; xaˆ; y xa − yl z2 − xm z3 − yn−1−i . a (λ1−1,λ˜) S = = ⎜ ( ( ) ) Ml 1 ( ) mM1 ( ) Mi 1 ( )⎟ = m=a = ⎝ ≠ ⎠

So the equality LHS z1 xa RHS z1 xa comes from Proposition 12.4, with the vari- S = = S = ables shifted by x → xaˆ and z → σ z . ( ) For b 2, assume the statement holds for b − 1 . Setting z1 ∶ xa on both sides, Proposition> 12.3 gives ( ) =

n−b+1 b−1 b+λ1−λ˜1 n−1 LHS z1 x S 1 z; xaˆ; y xa − yl zi − yn+1−b zi − xn+1−λ1 a ((λ1)b− ,λ˜) S = = ( ) Ml 1 ( ) Mi 2( ) iMb+2 ( ) = = = n−b+1 mul((λ1−1,λ˜))−1 n−1 RHS z1 x ∂b ⎛S 2 z; xaˆ; y xa − yl zb+1 − yn+1−b−i ⎞ . a ((λ1)b− ,λ1−1,λ˜) S = = ( ) Ml 1 ( ) Mi 1 ( ) ⎝ = = ⎠

The equality LHS z1 xa RHS z1 xa comes from the induction hypothesis with the S = = S = variables shifted by x → xaˆ and z → σ z . ( ) 48 DONGHYUNKIMANDLAURENK.WILLIAMS

References [AGS20] Erik Aas, Darij Grinberg, and Travis Scrimshaw. Multiline queues with spectral parameters. Comm. Math. Phys., 374(3):1743–1786, 2020. [AL14] Arvind Ayyer and Svante Linusson. An inhomogeneous multispecies TASEP on a ring. Adv. in Appl. Math., 57:21–43, 2014. [AL18] Erik Aas and Svante Linusson. Continuous multi-line queues and TASEP. Ann. Inst. Henri Poincaré D, 5(1):127–152, 2018. [AM13] Chikashi Arita and Kirone Mallick. Matrix product solution of an inhomogeneous multispecies TASEP. J. Phys. A, 46(8):085002, 11, 2013. [AS16] Erik Aas and Jonas Sjöstrand. A product formula for the TASEP on a ring. Random Structures Algorithms, 48(2):247–259, 2016. [BB93] Nantel Bergeron and Sara Billey. RC-graphs and Schubert polynomials. Experiment. Math., 2(4):257–269, 1993. [Can16] Luigi Cantini. Inhomogenous multispecies TASEP on a ring with spectral parameters, 2016. arXiv:1602.07921. [Can17] Luigi Cantini. Asymmetric simple exclusion process with open boundaries and Koorn- winder polynomials. Ann. Henri Poincaré, 18(4):1121–1151, 2017. [CdGW15] Luigi Cantini, Jan de Gier, and Michael Wheeler. Matrix product formula for Macdonald polynomials. J. Phys. A, 48(38):384001, 25, 2015. [CFF18] Rocco Chirivi’, Xin Fang, and Ghislain Fourier. Degenerate Schubert varieties in type A, 2018. arXiv:1808.01594. [CMW18] Sylvie Corteel, Olya Mandelshtam, and Lauren Williams. From multiline queues to Macdonald polynomials via the exclusion process, 2018. arXiv:1811.01024. [CW11] Sylvie Corteel and Lauren K. Williams. Tableaux combinatorics for the asymmetric exclusion process and Askey-Wilson polynomials. Duke Math. J., 159(3):385–415, 2011. [CW18] Sylvie Corteel and Lauren K. Williams. Macdonald-Koornwinder moments and the two- species exclusion process. Selecta Math. (N.S.), 24(3):2275–2317, 2018. [FI] OEIS Foundation Inc. (2019). The On-Line Encyclopedia of Integer Sequences. http://oeis.org. [FK96] Sergey Fomin and Anatol N. Kirillov. The Yang-Baxter equation, symmetric functions, and Schubert polynomials. In Proceedings of the 5th Conference on Formal Power Series and Algebraic Combinatorics (Florence, 1993), volume 153, pages 123–143, 1996. [FM07] Pablo A. Ferrari and James B. Martin. Stationary distributions of multi-type totally asymmetric exclusion processes. Ann. Probab., 35(3):807–832, 2007. [KM05] Allen Knutson and Ezra Miller. Gröbner geometry of Schubert polynomials. Ann. of Math. (2), 161(3):1245–1318, 2005. [LW12] Thomas Lam and Lauren Williams. A Markov chain on the symmetric group that is Schubert positive? Exp. Math., 21(2):189–192, 2012. [Man01] Laurent Manivel. Symmetric functions, Schubert polynomials and degeneracy loci, volume 6 of SMF/AMS Texts and Monographs. American Mathematical Society, Prov- idence, RI; Société Mathématique de France, Paris, 2001. Translated from the 1998 French original by John R. Swallow, Cours Spécialisés [Specialized Courses], 3. [Man17] Olya Mandelshtam. Toric tableaux and the inhomogeneous two-species TASEP on a ring. 2017. arXiv:1707.02663. [USW04] Masaru Uchiyama, Tomohiro Sasamoto, and Miki Wadati. Asymmetric simple exclusion process with open boundaries and Askey-Wilson polynomials. J. Phys. A, 37(18):4985– 5002, 2004.

Department of Mathematics, University of California, Berkeley, CA 94720 Email address: [email protected] Department of Mathematics, Harvard University, Cambridge, MA USA Email address: [email protected]