Page 1 of 1

Probabilistic Proofs of Combinatorial Identities*

Jessica Jay, University of Bristol

Probability, Analysis and Dynamics Conference 2022

*Based on joint work with M ́arton Bal ́azs and Dan Fretwell, arXiv:2011.05006 (2021)

Probabilistic Proofs of Combinatorial Identities*

Jessica Jay, University of Bristol

Probability, Analysis and Dynamics Conference 2022

*Based on joint work with M ́arton Bal ́azs and Dan Fretwell, arXiv:2011.05006 (2021)

Asymmetric Simple Exclusion

(ASEP)

Nearest neighbour interacting particle system on Z with at

most one particle per site:

LMP RMH Z

q

−1

q

0 < q < 1

↷X

Product Bernoulli stationary blocking measure:

μ

c

(η) = Y

i=−∞

q

−2(i−c)ηi

1 + q

−2(i−c)

.

Conserved Quantity: N(η) = P

i=1

(1 − ηi) −

P

0

i=−∞

ηi

Ergodic Decomposition of state space:

Ω = [

n∈Z

Ω

n where Ωn

:= {η ∈ Ω : N(η) = n}.

Conditional stationary measure:

ν

n

(η) = μ

c

(η|N(η) = n) =

P

m=−∞

q

m(m+1)−2mc

q

n(n+1)−2nc

· μ

c

(η).

Asymmetric Zero-Range (AZRP)

Nearest neighbour interacting particle system on Z<0 with

any number of particles per site and open right boundary:

LMP −1 0

q

−1

q

Product Geometric stationary blocking measure:

π(ω) =

Y

i=1

q

2ω−i

(1 − q

2i

).

ASEP-AZRP Correspondence

Exclusion and Zero-range are in correspondence (when

restricting to a given conserved quantity).

The number of holes between particle i and i + 1 in ASEP is

the number of particles at the site −i in AZRP

Jacobi Triple Product

Theorem.

X

m∈Z

q

m2

z

m

=

Y

i=1

(1 − q

2i

)(1 + q

2i−1

z)(1 + q

2i−1

z

−1

)

We can prove this by considering the ground states

of ASEP and AZRP

−5 −4 −3 −2 −1 0 1 2 3 4 5

Ground state η

0 ∈ Ω

0

of ASEP

−10−9 −8 −7 −6 −5 −4 −3 −2 −1 0

Ground state ω

0

of AZRP

Correspondence: ν

0

0

) = π(ω

0

), explicitly:

P

m∈Z

q

m2+(1−2c)m

Q

(1 + q

2i−1+(1−2c)

)(1 + q

2i−1−(1−2c)

)

=

Y

i=1

(1−q

2i

).

Letting z := q

1−2c

and rearranging completes the

proof. (Bal ́azs and Bowen 2018) □

Family of 0-1-2 Systems

Nearest neighbour interacting particle systems on Z with up to 2 particles per site:

LMP RMH Z

p(2,0)

q(0,1) ↷

↷X

Assumptions on rates:

•Asymmetric: the system has right drift

•Attractive: p(2, ·) ≥ p(1, ·) > 0 and p(·, 0) ≥ p(·, 1) > 0

q(0, ·) ≥ q(1, ·) > 0 and q(·, 2) ≥ q(·, 1) > 0

•Algebraic Conditions: p(1,0)

q(0,1) =

p(2,1)

q(1,2) and p(1,0)p(2,1)q(1,1)q(0,2)

q(0,1)q(1,2)p(2,0)p(1,1) = 1

Product Stationary Blocking Measure:

μ

c

(η) = Y

i=−∞

t

I{ηi=1}

q ̃

−(i−c)ηi

1 + tq ̃

−(i−c) + q ̃

−2(i−c)

Conserved Quantity: N(η) = P

i=1

(2 − ηi) −

P

0

i=−∞

ηi

Conditional stationary measure (n = 0):

ν

0

(η) =

2

P

l=−∞

q ̃

l(l+1)−2lcμ

c

(η)

1 + Q

i=−∞

(1 − 2μ

c

i

(1))

Corresponding Family

By considering the inter-particle distances in the original process we find a family of

restricted nearest neighbour interacting particle systems on Z<0 with any number of

particles per site and open right boundary:

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0

Restriction: No two consecutive empty sites!

Dynamics: The dynamics are inherited from the original process i.e. the jump rates can

be written down in terms of the original jump rate functions p(·, ·) and q(·, ·).

Product Stationary Measure:

π

e

(ω) =

q ̃

P

i odd

iω−i+

P

i even

i(ω−i−1)

t

2

P

i odd

I{ω−i≥1}−

P

i even

I{ω−i=0}

Seven(q, t ̃ )

Here we have focused on the states that come from standing up states with even conserved

quantity. We can do a similar thing with odd conserved quantity; this just shifts the circled

pattern above onto the odd sites.

A New Identity!

Ground states:

−5 −4 −3 −2 −1 0 1 2 3 4 5

η

e

−10 −9 −8 −7 −6 −5 −4 −3 −2 −1

ω

e

Using the correspondence we have that

ν

0

e

) = π

e

e

),

rearranging and letting z := q ̃

−c we have proved:

2

X

l∈Z

Seven(q, t ̃ )q ̃

l(l+1)z

2l =

Y

i≥1

(1 + tzq ̃

i + z

2

q ̃

2i

)(1 + tz−1

q ̃

i−1 + z

−2

q ̃

2(i−1))

+

Y

i≥1

(1 − tzq ̃

i + z

2

q ̃

2i

)(1 − tz−1

q ̃

i−1 + z

−2

q ̃

2(i−1))

Generalized Frobenius Partitions (GFP’s)

Frobenius partitions, of n, are two row arrays which correspond to ordinary partitions of n (see left picture below); these arrays

have strictly decreasing rows ai and bi and are such that n = s +

P

s

i=1

(ai + bi). The Jacobi Triple Product Identity gives an

equivalence of generating functions for ordinary partitions.

Partition of 30

Corresponding GFP of 30

7 6 4

6 3 1

4 2 2

2 2 0

Corresponding generalised

Young diagram

GFP (with repeats)

of 15

We can generalise the notion of Frobenius partition to two-rowed arrays with certain conditions on the rows. For example, we

take ai and bi to be non-increasing such that no part repeats more than once in a row. These no longer correspond to ordinary

partitions but can be represented by some generalised Young diagram (see right picture above). The identity found by considering

the 0-1-2 family gives an equivalence of generating functions for these GFP’s (the power of t gives the number of distinct parts).

t :=

p(1, 0)q(0, 2)

q(0, 1)p(1, 1)

1

2

≥ 1

q ̃ :=

q(0, 1)

p(1, 0) < 1

We must consider the cases

when n ∈ 2Z and n ∈ 2Z + 1

separately.

We always see this pattern far to the left.

Makes finding stationary distribution more complicated.

A similar identity is proved by considering the odd ground states. These

identities are new and specialise to give known 2-variable identities.