54
Ant foraging dynamics: From reaction-diffusion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon (U. Nice, INRIA) Instituto de Matem´ atica - Universidade Federal do Rio de Janeiro 46th John H. Barrett Memorial Lectures 2016 Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Embed Size (px)

Citation preview

Page 1: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Ant foraging dynamics: From reaction-diffusion toindividual-based models

Paulo AmorimRicardo Alonso (PUC Rio), Thierry Goudon (U. Nice, INRIA)

Instituto de Matematica - Universidade Federal do Rio de Janeiro

46th John H. Barrett Memorial Lectures 2016

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 2: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Leafcutter ant foraging trail

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 3: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

• Why study ants?

Image: www.alexanderwild.com (l), W.P. Armstrong (r).

• Besides scientific interest, some ant species like Solenopsis invicta(the Fire Ant, left) and Tapinoma melanocephalum (Ghost Ant,right) are considered invasive or pests.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 4: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Ant foraging

• One of the most interesting emergent behaviors in the animal kingdom• Individuals following simple, local, behavioral rules produce complex,“intelligent” large-scale strategies

• In general, ants have weak vision and very limited cognitive ability• How can they originate complex, organized collective behavior?

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 5: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Ant communication

• Ants use tactile, visual, chemical, and even acousticcommunication.

• One of the most important:pheromones.• Chemical compounds whichindividual ants secrete and leaveon the substrate.• Used to signal alarm, presenceof food, or colony-specific olfac-tory signatures.• The queen uses pheromonesto inhibit workers’ sexual devel-opment.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 6: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Foraging dynamics

• Goal: Efficient discovery and removal of food sources

• Key: Ants are attracted to foraging pheromone

• Pheromone laying produces a Feedback loop:

1 Ants find food through random foraging;2 Return to nest carrying food and depositing pheromone;3 Other ants detect pheromone and follow the trail to food source,

repeating the process;4 Trail is reinforced.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 7: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Modeling foraging

• We wish to find a mathematical framework where evolutionaccording to simple rules produces trail formation.

• Previous approaches: Mostly discrete models. Continuous modelsonly for isolated parts of the foraging dynamics.

• No “complete” model so far, incorporating the several stages offoraging.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 8: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Modeling animal movement

• Suppose the ant moves randomly on a 1-d territory.• The ant starts from x = 0 at t = 0.• p(x , t) = Probability of finding the ant at x at time t. Suppose in eachδt increment, it moves δx (to the left) or −δx (to the right). Then,

p(x , t) =1

2p(x − δx , t − δt) +

1

2p(x + δx , t − δt).

Or:

If the ant is ar x at time t, then either it was at x − δx or at x + δx attime t − δt, with equal probability.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 9: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

(∗) p(x , t) =1

2p(x − δx , t − δt) +

1

2p(x + δx , t − δt)

Taylor expansion:

p(x − δx , t − δt) = p(x , t)

− δx ∂p∂x− δt ∂p

∂t+ (δx)2∂

2p

∂x2+ δtδx

∂2p

∂x∂t+ (δt)2∂

2p

∂t2+ · · ·

p(x + δx , t − δt) = p(x , t)

+ δx∂p

∂x− δt ∂p

∂t+ (δx)2∂

2p

∂x2− δtδx ∂2p

∂x∂t+ (δt)2∂

2p

∂t2+ · · ·

Plugging into (∗), we get

∂p

∂t=

(δx)2

2δt

∂2p

∂x2+δt

2

∂2p

∂t2+ · · ·

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 10: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

∂p

∂t=

(δx)2

2δt

∂2p

∂x2+δt

2

∂2p

∂t2+ · · ·

• Say δt, δx → 0 in such a way that (δx)2

2δt → D ≡ constant. In the limitwe get

∂p

∂t= D

∂2p

∂x2,

which is just the diffusion equation. In n-d,

∂p

∂t= D∆p.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 11: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Reaction-diffusion-advection equation

More generally:

∂p

∂t− D∆p

↑Diffusion due to

random movement

+

Transport in the directionof a velocity vector V

↓div(pV) = F (p, x , t)

↑Source or sink

• p may be seen as the ant density

• Velocity V =??

• what if V is the concentration gradient of a chemical?

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 12: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Chemotaxis

• Foraging behavior falls into the scope of chemotaxis:

Chemotaxis system ∂tρ−∆ρ + div(χρ∇c) = 0

∂tc −∆c + τc = ρ

• Used for modeling aggregations of bacteria, slime molds, cells, etc.

• ρ is the density of bacteria

• c is the density of the chemoattractant

• The model contains: random movement of bacteria, production ofthe chemical by the bacteria, attraction to the chemical, evaporation,diffusion of the chemical.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 13: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Chemotaxis

∂tρ−∆ρ + div(ρχ∇c) = 0

∂tc −∆c + τc = ρ

• Some mathematical properties:

• Possible blow-up of solutions (depending on spatial dimension andinitial data)

• Aggregation

• Formation of spatio-temporal patterns

• Rich mathematical structure

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 14: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

A chemotactical model of ant foraging

• Ant population is represented by density functions of t > 0 andx ∈ Ω ⊂ R2

• Population is divided into two groups

• Foraging ants u(t, x)

• Returning ants w(t, x)

• Pheromone concentration v(t, x)

• Food distribution c(t, x)

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 15: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

A chemotactical model of ant foraging

• Simple dynamical rules:

• Foraging ants emerge from the nest and disperse randomly with atendency to follow the direction of increasing pheromone, if any;

• Once reaching food, foraging ants turn into returning ants andcollect food;

• Returning ants deposit pheromone as they return to the nest;

• Upon reaching the nest, returning ants turn into foraging ants.

• Pheromone diffuses and evaporates.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 16: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Modeling assumptions

• Returning to the nest

We prescribe a homebound vector field ∇a(x). Supported inliterature: ants use various non-pheromone methods to return to thenest (astromenotaxis, path integration, visual landmarks, CO2concentration, home range marking, etc...).

• Dynamics near nest and food

The nest is a special area where returning ants change into foragingants. We do not attempt to model other methods of recruitment nearthe nest.The food site is a special area where foraging ants turn into returningants, depleting the food.

• Experimental fact: pheromone deposition decreases as antsapproach the nest.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 17: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

A chemotactical model of ant foraging

∂tu − αu∆u + div

(u βu∇v

)= −λ1uc + λ2wN(x) + M(t)N(x)

∂tw − αw∆w + div(w βw∇a

)= λ1uc − λ2wN(x)

∂tv = µP(x)w − δv + αv∆v

∂tc = −γu c .

• Foraging ants u(t, x)

• Returning ants w(t, x)

• Pheromone concentration v(t, x)

• Food distribution c(t, x)

• M,N and P describe the nest, emergence of foraging ants from the nest,and decrease in pheromone deposition approaching the nest.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 18: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

A chemotactical model of ant foraging

• Nondimensional system

P.A., JTB 2015

∂tu − ∆u︸︷︷︸random dispersal

+

follow increasing pheromone︷ ︸︸ ︷div(u χu∇v

)= −uc + λwN(x) + M(t)N(x)︸ ︷︷ ︸

ant type change and emergence from nest

∂tw − Dw∆w + div(w ∇a

)︸ ︷︷ ︸follow home-bound field

= uc − λwN(x)

∂tv = P(x)w − εv + Dv∆v︸ ︷︷ ︸pheromone dynamics

∂tc = −u c︸ ︷︷ ︸food removal

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 19: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

No-flux Boundary conditions

(∇u − χuu∇v

)· n = 0(

Dw∇w − w∇a)· n = 0

∇v · n = 0, c = 0

• Conservation of total population:

∫Ωu(x) + w(x) dx = C ,

after all the ants have left the nest.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 20: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Numerical simulation

• Numerical difficulties: Sharp fronts, narrow trails with largegradients

• We use an upwind finite difference scheme for the spatial derivativesand a fourth order Runge–Kutta scheme for the time evolution

• Numerical results exhibit the main phenomenon: trail formationto and from the nest:

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 21: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Numerical results - foraging ants

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 22: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Numerical results - returning ants

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 23: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Numerical results - pheromone concentration

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 24: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Numerical results - food

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 25: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Numerical results - food mass evolution

Hours

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 26: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Numerical results - returning ants with topography

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 27: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Numerical results with topography

Returning ants Foraging antsPaulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 28: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Parameter space exploration: conditions for trail formation

• Evaporation rate ε vs. Chemotactic sensitivity χu

• Conditions for trail formation inforaging ants. A star ? indicatessimulations where, for fixed ε, thefood removal efficiency is greatest.A white star indicates simulationswhere, for fixed χu, the food re-moval efficiency is greatest.

• Values with marked trail formationcoincide with greater food removalefficiency.

Figure 5: Conditions for trail formation in foraging ants. From left to right, " =0.01, 0.1, 0.5, 1, 2.5, 5. Bottom to top, u = 20, 40, 80, 160, 500, 1000. " is the pheromonedegradation rate, u is the foraging ants’ chemotactic sensitivity (see Table 2). A star? indicates simulations where, for fixed ", the food removal efficiency is greatest. Awhite star indicates simulations where, for fixed u, the food removal efficiency isgreatest (see Table 4).

22

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 29: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Analytical results (with R. Alonso and Th. Goudon)

Fast pheromone diffusion ant foraging system∂tu −∆u + div

(u∇p

)= −uc + wN

∂tw −∆w + div(w ∇v

)= uc − wN

−∆p = w − p

• Pheromone diffusion is fast relative to ants’ movement

• Food source c(x) is very large or renewable: no depletion

• N(x) is the (given) nest function

• ∇v is the direction to the nest

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 30: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Theorem (R. Alonso, PA, Th. Goudon, to appear in M3AS)

Let T > 0, 0 ≤ (u0,w0) ∈ L1 ∩ L2+(Ω), Ω ⊂ R2. There exists a unique

nonnegative weak solution for the fast pheromone diffusion ant foragingsystem. For any 0 < t ≤ T :

(i) Lγ-integrability

‖w(t)‖γ + ‖u(t)‖γ ≤ C(

1 +1

t(1/γ′)+

), γ ∈ [1,∞) ,

(ii) L∞-integrability

‖w(t)‖∞ + ‖u(t)‖∞ ≤ C(

1 +1

t1+

).

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 31: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Proposition (a priori estimates)

Let (u,w) be a classical nonnegative solution of the fast pheromonediffusion ant foraging system. Then, for any γ ∈ [1,∞] we have theestimate

‖w(t)‖γ + ‖u(t)‖γ ≤ C(

1 +1

t(1/γ′)+

), t > 0 ,

with constant independent of time. Furthermore, for any γ ∈ [1,∞] thisestimate can be upgraded to∫

Ωwγ (t) dx +

∫Ωuγ(t) dx ≤ C (m0, ‖u0‖γ , ‖w0‖γ ) , t > 0 ,

if ‖u0‖γ , ‖w0‖γ <∞.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 32: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Why no blow-up?

• Usual chemotaxis system:∂tρ−∆ρ = − div(ρ∇c),

−∆c = ρ− c.

• − div(ρ∇c) = −∇ρ · ∇c − ρ∆c = −∇ρ · ∇c + ρ2 − ρc

• Formally,

∂tρ−∆u +∇ρ · ∇c + ρc = ρ2.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 33: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Why no blow-up?

∂tu −∆u + div

(u∇p

)= −uc + wN

∂tw −∆w + div(w ∇v

)= uc − wN

−∆p = w − p

• − div(u∇p

)= −∇u · ∇p − u∆p ≤ −∇u · ∇p + uw

• ∂tu −∆u +∇u · ∇p ≤ −uc + wN + uw

• But w is a solution to a parabolic equation; uw is better than u2.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 34: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

A priori estimates

• Tools: Gagliardo–Nirenberg–Sobolev inequality in 2-d:

∫Ωξα+1 ≤ C (Ω)‖ξ‖1

∫Ω|ξα|+ |∇ξα/2|2 , ∀α ≥ 1 ,

• Mass conservation: ∫Ωu(t) +

∫Ωw(t) = m0

• Lebesgue interpolation: r > 1,

‖u‖r ≤ ‖u‖1−θ1 ‖u‖θr+δ, θ ∈ (0, 1)

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 35: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

A priori estimates

• 1 < γ < α < 1 + γ <∞,

• d

dt

∫Ωuγ dx + C

∫Ωuγ+1 dx .

∫Ωuγ dx +

∫Ωwγ dx +

∫Ωwγ+1 dx ,

• d

dt

∫Ωwα dx + C

∫Ωwα+1 dx .

∫Ωuα dx +

∫Ωwα dx

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 36: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

A priori estimates

• 1 < γ < α < γ + 1 <∞

• If |Ω| <∞, use wγ+1 ≤ C + εwα+1, etc

• d

dt

∫Ωuγ dx + C

∫Ωuγ+1 dx . C + ε

∫Ωuγ+1 dx + ε

∫Ωwα+1 dx ,

• d

dt

∫Ωwα dx + C

∫Ωwα+1 dx . C + ε

∫Ωuγ+1 dx + ε

∫Ωwα+1 dx

• d

dt

∫Ωuγ + wα dx + C

∫Ωuγ+1 + wα+1 dx ≤ C

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 37: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

A priori estimates

• Use ‖u‖γ ≤ C‖u‖θγ+1 with the right θ to find

d

dtZ + C Z

αα−1 ≤ C

with

Z (t) =

∫Ωuγ +

∫Ωwα.

ODE comparison result

Z (t) ≤ C(

1 +1

tα−1

)which is essentially the desired result

‖w(t)‖γ + ‖u(t)‖γ ≤ C(

1 +1

t(1/γ′)+

), t > 0 .

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 38: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

L∞ estimate

‖w(t)‖∞ + ‖u(t)‖∞ ≤ C(

1 +1

t1+

)• L∞ estimates by a DiGiorgi energy method for the level sets (Moseriteration also works in some cases). Consider

∂tw −∆w +∇ · (w∇v) = f , in [0,T ]× Ω ,

• For M > 0 define the level sets and the sequence of levels and times

wk = (w − λk) 1w>λk ,

λk =(1− 1/2k

)M , tk =

(1− 1/2k+1

)t? , t? > 0.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 39: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

L∞ estimate

wk = (w − λk) 1w>λk

• Energy functional for the level sets

Wk := supt∈[tk ,T ]

∫Ωw2k dx +

∫ T

tk

∫Ω

∣∣∇wk

∣∣2 dx dt,

• We have ∂tw −∆w +∇ · (w∇v) = f ⇒

d

dt

∫Ωw2k dx +

∫Ω

∣∣∇wk

∣∣2 dx ≤ ∫Ω

∣∣w∇v ∣∣2 1w>λkdx + 2

∫Ωf+ w dx .

• Estimate the right-hand side using Wk−1.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 40: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

L∞ estimate

Note that if wk > 0 then wk−1 ≥ (λk − λk−1)M = 2−kM, and, as aconsequence,

1w>λk ≤(2k

Mwk−1

)a, ∀ a ≥ 0 . (1)

This and a Sobolev inequality allows to bound terms with wk using(wk−1)a.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 41: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

L∞ estimate

• Prove a recurrence relation Wk ≤ C (M, t?)W ak−1, a > 1, giving Wk → 0

• From

Wk ≥1

T − t?

∫ T

t?

∫Ωw2(t, x) 1w(t,x)≥M(1−1/2k ) dx dt

we conclude that ∫ T

t?

∫Ωw2(t, x) 1w(t,x)≥M dx dt = 0,

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 42: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

L∞ estimate

and so0 ≤ w(t, x) ≤ M a. e. (t?,T )× Ω.

Moreover, M ≤ C1

t1+?

giving

‖w(t)‖∞ + ‖u(t)‖∞ ≤ C(

1 +1

t1+

).

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 43: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Stability, uniqueness

• Existence of classical short-time solution follows by a contractionargument, which can be extended to t =∞ by the a priori estimates.

Stability

Given two initial data (ui ,o ,wi ,o) ∈ L1 ∩ L2+(Ω), with i = 1, 2, then

solutions (ui ,wi ) satisfy the estimate

‖u1(t)− u2(t)‖2 + ‖w1(t)− w2(t)‖2

≤(‖u1,o − u2,o‖2 + ‖w1,o − w2,o‖2

)eC(mo)t , t ≥ 0 .

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 44: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Stability, uniqueness

• u = u1 − u2, ...

• Use equations and previous estimates to get

d

dt‖w‖2

2 + ‖∇w‖22 ≤ C

(‖w‖2

2 + ‖u‖22

),

andd

dt‖u‖2

2 +1

2‖∇u‖2

2 ≤ C(

1 +1√t

)(‖u‖2

2 + ‖w‖22

).

• With Z (t) := ‖u‖22 + ‖w‖2

2,

dZ (t)

dt≤ C

(1 +

1√t

)Z (t) .

• Integrate to get the stability result.

• Allows to construct weak solution by approximation.Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 45: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

An Individual-Based Model of ant trail following

“The exact nature of how individual ants move on the trail andrespond to pheromone remains unknown.” A. Perna et al. (2012) Individual

Rules for Trail Pattern Formation in Argentine Ants (Linepithema humile). PLoS Comput Biol 8(7)

• At microscopic level, the continuous model only considers attractiontowards higher concentrations of pheromone.

• Trail-following must involve something more, since in nature trails donot have pheromone gradient along the trail itself.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 46: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

An Individual-Based Model of ant trail following

• Reaction-diffusion models are at one end of a modeling spectrum:macroscopic models.

• At the other end are discrete or individual-based (IBM) models.

• Individual ants are modeled by the law of motiond

dtX (t) = V (t),

d

dtV (t) = −1

τ

(V − F (X ,V )

)• τ is a relaxation time and F (X ,V ) is the desired velocity.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 47: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

An Individual-Based Model of ant trail following

• Traditional approach: the desired velocity F is proportional to thegradient of an attracting chemical.

• Our assumption: Ants are attracted to the presence of pheromone on acircular sector representing sensing area.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 48: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

An Individual-Based Model of ant trail following

• Sensing area:

B(V , `, θ∗) = X ∈ R2 : ∠(V ,X ) ∈ (−θ∗, θ∗), ‖X‖ ≤ `.

• Desired velocity:

F (X ,V ) =

∫B(V ,`,θ∗)

X ′c(t,X + X ′) dX ′

• c(t,X ) is the pheromone concentration.

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 49: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

An Individual-Based Model of ant trail following

Weber’s Law:

Individual response ' L− R

L + R,

where L is stimulus detected on left side, and R on right side.

F (X ,V ) = λ

∫BX ′c(t,X + X ′) dX ′∫Bc(t,X + X ′) dX ′

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 50: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

An Individual-Based Model of ant trail following

• How to remove the singularity of F (X ,V ) = λ

∫B X ′c(t,X + X ′) dX ′∫B c(t,X + X ′) dX ′

when the pheromone is zero, and introduce a pheromone detectionthreshold?

• Ants move at near-constant speed.

• Pheromone below a small threshold c∗ is not detected.

• With constant pheromone, the desired velocity F (X ,V ) simplifies to thenatural velocity

CV

‖V ‖.

⇒ Replacing c(t,X ) with c(t,X ) := max(c(t,X ), c∗) removes thesingularity and makes the ant move at the natural velocity whenpheromone is below threshold.

“pheromone below the threshold is treated as constant pheromone”Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 51: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

A linearized model

F (X ,V ) = λ

∫BX ′c(t,X + X ′) dX ′∫Bc(t,X + X ′) dX ′

.

Taylor develop: c(t,X + X ′) ' c(t,X ) + X ′ · ∇c(t,X ). Then

Flin(X ,V ) = λ

c(t,X )

∫B

X ′ dX ′ +

∫B

X ′ ⊗ X ′ dX ′∇c(t,X )

c(t,X )|B|+∇c(t,X ) ·∫B

X ′ dX ′

= λc(t,X )`

2

3sin θ∗ V

‖V‖ +`2

4

(θ∗I +

1

2sin 2θ∗A(Θ)

)∇c(t,X )

c(t,X )θ∗ +∇c(t,X ) · `2

3sin θ∗ V

‖V‖

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 52: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

A linearized model

• Θ is the angle of the ant’s current velocity, V . So the (linearized)desired velocity is

Flin(X ,V ) = λc(t,X )`

2

3sin θ∗ V

‖V ‖ +`2

4

(θ∗I +

1

2sin 2θ∗A(Θ)

)∇c(t,X )

c(t,X )θ∗ +∇c(t,X ) · `2

3sin θ∗ V

‖V ‖

with

A(θ∗) =

(cos 2Θ sin 2Θ

sin 2Θ −cos 2Θ

),

V = ‖V ‖(cos Θ, sin Θ).

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 53: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Some results

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics

Page 54: Ant foraging dynamics: From reaction-diffusion to ... · Ant foraging dynamics: From reaction-di usion to individual-based models Paulo Amorim Ricardo Alonso (PUC Rio), Thierry Goudon

Thank you for your attention!

http://www.im.ufrj.br/paulo/

Paulo Amorim - UFRJ - Rio de Janeiro Ant Foraging Dynamics