Research Statement Peyam Ryan Tabrizian October 2015

1 Past and current research

Introduction My research lies primarily in the field of partial differential equations (PDE)
and the calculus of variations. I am interested in studying various asymptotic limits that
arise in PDE, such as in chemical reactions. More precisely, consider the following one-
dimensional Kramers-Smoluchowski equation:

τ (ρt − a∆x ρ ) = ρξ + −2 ρ Φ0 (ξ)

ξ
(KS )
1
Here ρ = ρ (x, ξ, t) : U ×R×[0, T ] → R (U open and bounded in Rn ), τ = 12 e− 2 , a = a(ξ)
is bounded and positive, Φ : R → R, Φ = Φ(ξ) is an even double-well potential function as
depicted in Figure (a), normalized so that Φ(0) = 1, Φ(±1) = 0, Φ(±2) = 1:

(a) Double-Well Potential (b) Triple-Well Potential (c) Infinitely many Wells

−Φ
e2
´ ρ
By defining σ  = Z , where Z makes Rσ
 = 1, and letting u =: σ , we can normalize
(KS ) to become:
0
τ σ  (ut − a ∆x u ) = σ  uξ


ξ
(KS )
Chemical Context There are two different ways of viewing a chemical reaction. Consider
a simple A * ) B-system, where a particle A reacts to become B, and vice-versa. On the
other hand, the dynamics of the reaction are given by the classical reaction-diffusion system
(R−D) (see below), where α is the density of A and β is the density of B. On the other
hand, one can augment the system by adding a ‘chemical’ variable ξ, so that a chemical
reaction corresponds to the movement of a particle from one local minimum (here ξ = −1,
corresponding to A) to another (ξ = 1, corresponding to B). In that case, the dynamics of
the reaction are given by an SDE, whose Fokker-Planck equation is (KS ). The main result
below shows that both points of view are just different sides of the same coin: one can take
a limit of large activation energy 1 of (KS ) to obtain (R−D).

Main Result Using standard estimates, one can show that:

Theorem. For all 0 ≤ t ≤ T , we have:
ρ * α δ{ξ=−1} + β δ{ξ=1}

1
Research Statement Peyam Ryan Tabrizian October 2015

where α = α(x, t) and β = β(x, t); and more importantly:

Theorem (Main Theorem). The √ functions α and β solve the following linear reaction-
|Φ00 (0)|Φ00 (1)
diffusion system, where κ = 2π and d± := a(±1):
(
αt − d− ∆α =κ(β − α)
(R−D)
βt − d+ ∆β =κ(α − β)

Idea of proof This result has already been proven by Peletier, Savaré, and Veneroni in
[PSV12], using Γ−convergence. In [HN11], Herrmann and Niethammer provide a different
proof, by rewriting (KS ) as a gradient flow on the Wasserstein space of probability measures
and using a Rayleigh-type dissipation functional. In my thesis [Tab16] and in a joint paper
with my advisor Lawrence C. Evans [ET16], we provide a direct proof that avoids the use of
abstract machinery. The main idea is to devise a test function φ which, after multiplying
0
(KS ) by φ and integrating by parts, cancels out the singular term σ  in (KS ):

ˆ Λ(ξ) −3/2 if s ≤ −3/2

 τ
φ (ξ) = dξ where Λ(s) = s if − 3/2 ≤ s ≤ 3/2
0 σ 
3/2 if s ≥ 3/2


The proof is robust enough that we can modify it to treat more general cases. All proofs
rely on building test functions similar to φ above.

Generalizations:

Three wells: If Φ has three wells at −2, 0, 2 as in figure (b), then ρ * α δ−2 + β δ0 + γ δ2 ,
where α, β, δ solve (Here di = a(i)):

αt − d−2 ∆x α = κ (β − α)

βt − d0 ∆x β = κ (α − 2β + γ) (R-D)

γt − d2 ∆x γ = κ (β − γ)


Periodic wells: Take the triple-well case, but this time identify the points − 52 and 72 and
´ 7/2
modify σ  so that −5/2 σ  = 1. Then we get the same result as for the triple-well-case.

Infinitely
´ 1 many wells: If Φ(2m) = 0P for m ∈ Z, as in Figure (c), then modifying Z
so that −1 σ  dξ = 1, we get that ρ * ∞ m
m=−∞ α δ2m for functions α
m (m ∈ Z), which

satisfy the infinite system, where d2m =: a(2m):

αtm − d2m ∆x αm = 2κ αm−1 − 2αm + αm+1

2
Research Statement Peyam Ryan Tabrizian October 2015

Higher-dimensional case: In the joint paper above [ET16], we were able to generalize
this to the case where the chemical variable ξ is more than one-dimensional. Assume that
Φ : Rm → R is smooth, nonnegative, even in the first variable ξ1 , has two wells at the points
e± = (±1, 0, · · · , 0), normalized so that Φ(0) = 1, Φ(e± ) = 0, and moreover det D2 Φ(e± ) 6= 0
and D2 Φ(0) is diagonal, with eigenvalues λ1 (0) < 0 < λ2 (0) ≤ · · · ≤ λm (0). Then the analog
0
of (KS ) reads as:

τ σ  (ut − a∆x u ) = divξ (σ  Dξ u )

And we obtain that ρ * αδe− + βδe+ , where α, β solve:
(
αt − d− ∆x α = κ (β − α)
βt − d+ ∆x β = κ (α − β)
√
|det D2 Φ(e± )|
Here d± = a(e± ) and κ = |λ12π(0)| √ 2
|det D Φ(0)|
In this case, to calculate κ, we use capacity-methods, and to construct our
 test-function φ ,
±
we show that there exists a solution to the following PDE, where B =: Φ(ξ) ≤ 4 ∩ Rm 1

±:
  
σ  1 1
− div Dφ = + χB+ − − χB−
τ |B | |B |

2 Future and Planned Research

In my postdoctoral studies, I intend to study more general asymptotic PDE models, includ-
ing but not limited to chemical reactions. The following is a list of possible research topics
that I am interested in.

Nonlinear Chemical Reactions and Diffusions One idea is to work on an extension

of the chemical reactions and diffusions-paper: Since I am interested in nonlinear PDE,
I ultimately would like to find an equation similar to (KS ) with the property that the
solutions ρ * αδ−1 + βδ1 , but this time α and β solve a nonlinear reaction-diffusion
system, where ψ i are nonlinear functions of α, β:
(
αt − d− ∆x α = ψ 1 (α, β)
αt − d+ ∆x β = ψ 2 (α, β)
One approach would be to work backwards in the direct proof I´provided, and to see how,
after cancellation with the test-function φ , instead of getting (u )ξ , one can obtain an
´
expression of the form (ψ(u ))ξ dξ for some suitable nonlinear ψ.

3
Research Statement Peyam Ryan Tabrizian October 2015

A Fisher-KPP-system I am also interested in reaction-diffusions systems arising in Bi-

ology, in particular the works of Henri Berestycki. In their paper [BRR13], Berestycki,
Roquejoffre, and Rossi study a system modeling the interaction between two populations,
one u = u(x, t) on a road and another one v = v(x, y, t) in a field (x ∈ R, y ∈ R+ ), where f
is of Fisher-KPP-type, meaning that f (0) = f (1) = 0 and 0 < f (s) ≤ f 0 (0)s on (0, 1):

ut − Duxx = νv(x, 0, t) − µu

vt − dvxx − dvyy = f (v)

−dvy (x, 0, t) = µu − νv(x, 0, t)


The authors show that there is an asymptotic speed of propagation (ASP) c∗ = c∗ (µ, ν, d, D) >
0 for which:

• For all c > c∗ , limt→∞ sup|x|≥ct (u(x, t), v(x, y, t)) = (0, 0)
 
• For all c < c∗ , limt→∞ sup|x|≤ct (u(x, t), v(x, y, t)) = µν , 1

I would like to study to what extent the above result can be refined. What happens to u
and v if we replace the region {(x, y) ∈ R × R≥0 , |x| ≤ ct} by the region, say, |x| ≤ c log(t)?
Would we get an even larger ASP there? Would the convergence be exponentially fast?
It is also shown that if D > 2d, then e c(µ, ν, d) =: limD→∞ c∗ (µ,ν,d,D)
√
D
exists. How does e c
depend on the parameters µ, ν, d?

A homogenization model for motor proteins Staying within the realm of Biology, I
am also interested in studying a nonlinear version of the following Fokker-Planck system an-
alyzed in the paper [MS13] by Souganidis and Mirrahimi (where µ, ν, ψ > 0 are 1−periodic),
which models the motion of motor proteins along molecular filaments:
(
ut − ∆x u − divx u Dy ψ x = 1 ν x v  − µ x u

vt − ∆x v  = 1 µ x u − ν x v 

In the linear case, the authors show that the

´ proteins move along a fixed filament with a
constant speed v ∈ R , where I0 =: lim→0 Rd u (x, 0) + v  (x, 0)dx:
d 

u (x, t) + v  (x, t) * δ(x − vt)I0

I would like to analyze to what extent this result still holds in the nonlinear case, where Φ
and Ψ are suitable nonlinear functions of their arguments:
(
ut − ∆x u − divx u Dy ψ x = 1 Φ u , v  , µ x , ν x

vt − ∆x v  = 1 Ψ u , v  , µ x , ν x

4
Research Statement Peyam Ryan Tabrizian October 2015

The G-equation Finally, inspired by the works of Jack Xin and Yifeng Yu on combustion
theory [XY14], I would like to study the following G−equation, a Hamilton-Jacobi equation
modeling turbulent combustion, in the case where n = 3 with V (x, y, z) = (C cos(y) +
A sin(z), B sin(x) + A cos(z), B cos(x) + C sin(y)), the chaotic Arnold-Beltrami-Childress
flow:
(
Gt + AV (x) · DG + |DG| = 0
G(x, 0) = p · x

Numerical simulations show that the turbulent flame speed sT (p, A) =: limt→∞ −G(x,t) t
grows linearly with A as A → ∞ (p is any unit vector in R3 ); I am proposing to prove this
claim rigorously.

References
[BRR13] Henri Berestycki, Jean-Michel Roquejoffre, and Luca Rossi. The influence of a line
with fast diffusion on Fisher-KPP propagation. J. Math. Biol., 66(4-5):743–766,
2013.

[ET16] L. Evans and P. Tabrizian. Asymptotics for the kramers-smoluchowski equation.

to appear, 2016.

[HN11] Michael Herrmann and Barbara Niethammer. Kramers’ formula for chemical
reactions in the context of Wasserstein gradient flows. Commun. Math. Sci.,
9(2):623–635, 2011.

[MS13] S. Mirrahimi and P. E. Souganidis. A homogenization approach for the motion of

motor proteins. NoDEA Nonlinear Differential Equations Appl., 20(1):129–147,
2013.

[PSV12] Mark A. Peletier, Giuseppe Savaré, and Marco Veneroni. Chemical reactions as
Γ-limit of diffusion [revised reprint of mr2679596]. SIAM Rev., 54(2):327–352,
2012.

[Tab16] Peyam R. Tabrizian. Asymptotic PDE Models for Chemical Reactions. PhD thesis,
University of California - Berkeley, 5 2016. to appear.

[XY14] Jack Xin and Yifeng Yu. Asymptotic growth rates and strong bending of turbulent
flame speeds of G-equation in steady two-dimensional incompressible periodic
flows. SIAM J. Math. Anal., 46(4):2444–2467, 2014.