Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions A directed polymer approach to the once-oriented last passage site percolation time constant in high dimensions Gregory J. Morrow Department of Mathematics University of Colorado at Colorado Springs Conference in Memory of Walter V. Philipp June, 2009 Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Directed path for d = 1. There are (2d)n directed paths of length n, with height of path y ∈ Zd . · y ∈ Z1 · · @ @· @ @ · · @ @· @ @· · %@ @ & @ @ · ·· %· & @ & @· @ @· @· @ @ @ @· @· @ @· @ @· @ @· @ @· t Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions “Point to Line” Last Passage Times I Nearest Neighbor “heights” in Zd : γt , t = 0, 1, 2, . . . , n (γ(0) = 0). Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions “Point to Line” Last Passage Times I Nearest Neighbor “heights” in Zd : I γt , t = 0, 1, 2, . . . , n (γ(0) = 0). − Directed Path: → γn := (t, γt ), t = 0, 1, 2, . . . , n. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions “Point to Line” Last Passage Times I I I Nearest Neighbor “heights” in Zd : γt , t = 0, 1, 2, . . . , n (γ(0) = 0). − Directed Path: → γn := (t, γt ), t = 0, 1, 2, . . . , n. I.I.D. Weights, η, at sites (t, y ), with E(|η|) < ∞ and (WLOG) Eη ≥ 0. Also called “passage times”, η(t, y ). Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions “Point to Line” Last Passage Times I I I I Nearest Neighbor “heights” in Zd : γt , t = 0, 1, 2, . . . , n (γ(0) = 0). − Directed Path: → γn := (t, γt ), t = 0, 1, 2, . . . , n. I.I.D. Weights, η, at sites (t, y ), with E(|η|) < ∞ and (WLOG) Eη ≥ 0. Also called “passage times”, η(t, y ). “Passage Time” along a directed path: n X → − T (γn ) := η(t, γt ). t=1 Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions “Point to Line” Last Passage Times I I I I Nearest Neighbor “heights” in Zd : γt , t = 0, 1, 2, . . . , n (γ(0) = 0). − Directed Path: → γn := (t, γt ), t = 0, 1, 2, . . . , n. I.I.D. Weights, η, at sites (t, y ), with E(|η|) < ∞ and (WLOG) Eη ≥ 0. Also called “passage times”, η(t, y ). “Passage Time” along a directed path: n X → − T (γn ) := η(t, γt ). t=1 I “Point to Point” Last Passage Time: − e0,n := max T (→ T γn ). γ0 =γn =0 Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions “Point to Line” Last Passage Time Constant I Superadditivity of “Point to Point” Last Passage Times: e0,n+m ≥ T e0,m + T em,m+n , so, by Kingman, T e0,n /n = µd , P-a.s. and in L1 . lim T n→∞ Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions “Point to Line” Last Passage Time Constant I Superadditivity of “Point to Point” Last Passage Times: e0,n+m ≥ T e0,m + T em,m+n , so, by Kingman, T e0,n /n = µd , P-a.s. and in L1 . lim T n→∞ I “Point to Line” Last Passage Times:P e 0,n := maxγ =0 n η(t, γt ). H 0 t=1 e 0,n } is a superadditive sequence : EH e 0,n+m ≥ EH e 0,n +EH e 0,m . {EH Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions “Point to Line” Last Passage Time Constant I Superadditivity of “Point to Point” Last Passage Times: e0,n+m ≥ T e0,m + T em,m+n , so, by Kingman, T e0,n /n = µd , P-a.s. and in L1 . lim T n→∞ I “Point to Line” Last Passage Times:P e 0,n := maxγ =0 n η(t, γt ). H 0 t=1 e 0,n } is a superadditive sequence : EH e 0,n+m ≥ EH e 0,n +EH e 0,m . {EH I “Point to Line” Last Passage Time Constant: e 0,n /n = sup EH e 0,n /n. νd = νd (η) := lim EH n→∞ n . Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions “Point to Line” Last Passage Time Constant I Superadditivity of “Point to Point” Last Passage Times: e0,n+m ≥ T e0,m + T em,m+n , so, by Kingman, T e0,n /n = µd , P-a.s. and in L1 . lim T n→∞ I “Point to Line” Last Passage Times:P e 0,n := maxγ =0 n η(t, γt ). H 0 t=1 e 0,n } is a superadditive sequence : EH e 0,n+m ≥ EH e 0,n +EH e 0,m . {EH I “Point to Line” Last Passage Time Constant: e 0,n /n = sup EH e 0,n /n. νd = νd (η) := lim EH n→∞ I n . By Hammersley and Welsh: νd = µd . Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Evaluate ν1 for Exponential and Geometric cases I If η is unit exponential then: ν1 (η) = 2. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Evaluate ν1 for Exponential and Geometric cases I If η is unit exponential then: ν1 (η) = 2. I If η is geometric (p), then: p ν1 (η) = (1 + 1 − p)/p. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Evaluate ν1 for Exponential and Geometric cases I If η is unit exponential then: ν1 (η) = 2. I If η is geometric (p), then: p ν1 (η) = (1 + 1 − p)/p. I For both these special cases a Shape Theorem holds: e0,(nx,ny ) = g (x, y ) lim (1/n)T n→∞ Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Evaluate ν1 for Exponential and Geometric cases I If η is unit exponential then: ν1 (η) = 2. I If η is geometric (p), then: p ν1 (η) = (1 + 1 − p)/p. I For both these special cases a Shape Theorem holds: e0,(nx,ny ) = g (x, y ) lim (1/n)T n→∞ I e0,n − nν1 )/n1/3 converges in Also in these special cases (T distribution to the Tracy-Widom law F2 [ref. K. Johansson CMP (2000), relations to random matrix theory] Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Directed Polymer: Partition Function and Free Energy I Directed paths are polymers. Higher total weights are favored. Partition function: X − Zn (β) := (2d)−n exp(βT (→ γn )). γ Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Directed Polymer: Partition Function and Free Energy I Directed paths are polymers. Higher total weights are favored. Partition function: X − Zn (β) := (2d)−n exp(βT (→ γn )). γ I − − Gibbs measure: µn (→ γ ) := (2d)−n exp(βT (→ γn ))/Zn (β). Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Directed Polymer: Partition Function and Free Energy I Directed paths are polymers. Higher total weights are favored. Partition function: X − Zn (β) := (2d)−n exp(βT (→ γn )). γ I I − − Gibbs measure: µn (→ γ ) := (2d)−n exp(βT (→ γn ))/Zn (β). By Jensen’s inequality, {E ln Zn } is superadditive; so define the Free Energy: f (β) := lim (1/n)E ln Zn = sup(1/n)E ln Zn n→∞ n Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Directed Polymer: Partition Function and Free Energy I Directed paths are polymers. Higher total weights are favored. Partition function: X − Zn (β) := (2d)−n exp(βT (→ γn )). γ I I − − Gibbs measure: µn (→ γ ) := (2d)−n exp(βT (→ γn ))/Zn (β). By Jensen’s inequality, {E ln Zn } is superadditive; so define the Free Energy: f (β) := lim (1/n)E ln Zn = sup(1/n)E ln Zn n→∞ I n Log Moment Generating function: λ(β) := ln E exp(βη) Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Directed Polymer: Partition Function and Free Energy I Directed paths are polymers. Higher total weights are favored. Partition function: X − Zn (β) := (2d)−n exp(βT (→ γn )). γ I I − − Gibbs measure: µn (→ γ ) := (2d)−n exp(βT (→ γn ))/Zn (β). By Jensen’s inequality, {E ln Zn } is superadditive; so define the Free Energy: f (β) := lim (1/n)E ln Zn = sup(1/n)E ln Zn n→∞ I I n Log Moment Generating function: λ(β) := ln E exp(βη) By Jensen’s inequality w.r.t. E again, f (β) ≤ λ(β). Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Properties of f (β) I fn (β) := E(ln Zn )/n is convex, because fn00 (β) is a variance relative to Gibbs measure. Therefore f (β) is convex. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Properties of f (β) I fn (β) := E(ln Zn )/n is convex, because fn00 (β) is a variance relative to Gibbs measure. Therefore f (β) is convex. I Further, since E(η) ≥ 0, fn (β) is non-decreasing. Indeed, for β1 > β0 , fn (β1 ) = fn (β0 )+ (1/n)E ln E exp(∆βT ) [exp(β0 T )/E exp(β0 T )] e exp(∆βT ) ≥ fn (β0 )+EE(∆β)T e = fn (β0 )+(1/n)E ln E /n ≥ fn (β0 ), where at the last step Jensen’s inequality was applied w.r.t. e Here E is the uniform measure on directed paths and E e is E. simply the Gibbs measure expectation. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Proposition 1 I Estimate P γ − exp(βT (→ γn )) from below by the maximum term: E(ln Zn ) = E ln[(2d)−n X − e 0,n − n ln(2d) exp(βT (→ γn ))] ≥ βEH γ Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Proposition 1 I Estimate P γ − exp(βT (→ γn )) from below by the maximum term: E(ln Zn ) = E ln[(2d)−n X − e 0,n − n ln(2d) exp(βT (→ γn ))] ≥ βEH γ I Estimate also the average above by the largest term: − e 0,n = max(βT (→ βEH γn )) ≥ E(ln Zn ). γ Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Proposition 1 I Estimate P γ − exp(βT (→ γn )) from below by the maximum term: E(ln Zn ) = E ln[(2d)−n X − e 0,n − n ln(2d) exp(βT (→ γn ))] ≥ βEH γ I Estimate also the average above by the largest term: − e 0,n = max(βT (→ βEH γn )) ≥ E(ln Zn ). γ I By dividing by n and taking n → ∞, we have thus proved: Proposition 1: f (β) ≥ βνd − ln(2d), and lim f (β)/β = νd . β→∞ Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Prototypical examples I Exponential case: λ(β) = − ln(1 − β) ≥ 0, 0 ≤ β < 1. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Prototypical examples I Exponential case: λ(β) = − ln(1 − β) ≥ 0, 0 ≤ β < 1. I Poisson case: λ(β) = exp(β) − 1 ≥ 0, 0 ≤ β < ∞. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Prototypical examples I Exponential case: λ(β) = − ln(1 − β) ≥ 0, 0 ≤ β < 1. I Poisson case: λ(β) = exp(β) − 1 ≥ 0, 0 ≤ β < ∞. I Gaussian case: λ(β) = β 2 /2, 0 ≤ β < ∞. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Prototypical examples I Exponential case: λ(β) = − ln(1 − β) ≥ 0, 0 ≤ β < 1. I Poisson case: λ(β) = exp(β) − 1 ≥ 0, 0 ≤ β < ∞. I Gaussian case: λ(β) = β 2 /2, 0 ≤ β < ∞. I Generic case: P(η > x) = exp(−x · x α ), x ≥ 0, some α > 0. λ(β) ∼ Cα β 1+1/α . Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Prototypical examples I Exponential case: λ(β) = − ln(1 − β) ≥ 0, 0 ≤ β < 1. I Poisson case: λ(β) = exp(β) − 1 ≥ 0, 0 ≤ β < ∞. I Gaussian case: λ(β) = β 2 /2, 0 ≤ β < ∞. I Generic case: P(η > x) = exp(−x · x α ), x ≥ 0, some α > 0. λ(β) ∼ Cα β 1+1/α . I When λ(β) exists for all β we know by the martingale theory first developed by Bolthausen CMP(1989), and later by Comets-Yoshida AP (2006) that for high d, λ(β) = f (β) for small β > 0. We essentially determine in a generic case for high d where the departure of the two curves λ(β) and f (β) occurs in terms of d. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions The Gaussian case I Carmona and Hu PTRF (2002) have shown in the Gaussian case (λ(β) = β 2 /2) that : p p f (β) ≤ min β 2 /2, β 2 ln(2d) , so, νd ≤ 2 ln(2d) Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions The Gaussian case I Carmona and Hu PTRF (2002) have shown in the Gaussian case (λ(β) = β 2 /2) that : p p f (β) ≤ min β 2 /2, β 2 ln(2d) , so, νd ≤ 2 ln(2d) I Obviously, putting md := E max(η1 , . . . , η2d ), e 0,n ≥ EH e 0,1 = md . νd = sup(1/n)EH n≥1 Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions The Gaussian case I Carmona and Hu PTRF (2002) have shown in the Gaussian case (λ(β) = β 2 /2) that : p p f (β) ≤ min β 2 /2, β 2 ln(2d) , so, νd ≤ 2 ln(2d) I Obviously, putting md := E max(η1 , . . . , η2d ), e 0,n ≥ EH e 0,1 = md . νd = sup(1/n)EH n≥1 I But one can show directly in the Gaussian case that p md = 2 ln(2d) + o(1) p Hence in this case: νd = 2 ln(2d) + o(1), as d → ∞. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Exponential case: νd & ln(2d), as d → ∞ I We have νd ≥ md . We first estimate md := E max(η1 , . . . , η2d ) from below. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Exponential case: νd & ln(2d), as d → ∞ I I We have νd ≥ md . We first estimate md := E max(η1 , . . . , η2d ) from below. Write G (t) := t 2d for t = t(x) := P(η ≤ x) = 1 − exp(−x). Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Exponential case: νd & ln(2d), as d → ∞ I I I We have νd ≥ md . We first estimate md := E max(η1 , . . . , η2d ) from below. Write G (t) := t 2d for t = t(x) := P(η ≤ x) = 1 − exp(−x). By the change of variables t = t(x) we have Z ∞ Z 1 2d md = xdt (x) = − ln(1 − t)dG (t). 0 0 Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Exponential case: νd & ln(2d), as d → ∞ I I I We have νd ≥ md . We first estimate md := E max(η1 , . . . , η2d ) from below. Write G (t) := t 2d for t = t(x) := P(η ≤ x) = 1 − exp(−x). By the change of variables t = t(x) we have Z ∞ Z 1 2d md = xdt (x) = − ln(1 − t)dG (t). 0 I 0 But, − ln is a convex function. So by Jensen’s inequality, Z 1 md ≥ − ln( (1 − t)dG (t)) = − ln(1 − 2d/(2d + 1)) 0 = ln(2d) + o(1), as d → ∞. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Exponential case: νd . ln(2d), as d → ∞ I Next we apply Proposition 1: λ(β) ≥ f (β) ≥ βνd − ln(2d). Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Exponential case: νd . ln(2d), as d → ∞ I Next we apply Proposition 1: λ(β) ≥ f (β) ≥ βνd − ln(2d). I Now λ(β) = − ln(1 − β), β < 1. Choose β = 1 − 1/ ln(2d). We get that ln ln(2d) ≥ νd (1 − 1/ ln(2d)) − ln(2d). Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Exponential case: νd . ln(2d), as d → ∞ I Next we apply Proposition 1: λ(β) ≥ f (β) ≥ βνd − ln(2d). I Now λ(β) = − ln(1 − β), β < 1. Choose β = 1 − 1/ ln(2d). We get that ln ln(2d) ≥ νd (1 − 1/ ln(2d)) − ln(2d). I It follows that νd ≤ ln(2d) + O(ln ln(2d)) = ln(2d)(1 + o(1)), as d → ∞. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Exponential case: νd . ln(2d), as d → ∞ I Next we apply Proposition 1: λ(β) ≥ f (β) ≥ βνd − ln(2d). I Now λ(β) = − ln(1 − β), β < 1. Choose β = 1 − 1/ ln(2d). We get that ln ln(2d) ≥ νd (1 − 1/ ln(2d)) − ln(2d). I It follows that νd ≤ ln(2d) + O(ln ln(2d)) = ln(2d)(1 + o(1)), as d → ∞. I By the previous page we therefore have νd ∼ ln(2d), as d → ∞. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Assumptions to Construct a Lower Bound for md I Write F (x) := P(η ≤ x). Assume F (x) = 1−exp(−u(x)), for u(x) non-decreasing to ∞, as x → ∞. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Assumptions to Construct a Lower Bound for md I Write F (x) := P(η ≤ x). Assume F (x) = 1−exp(−u(x)), for u(x) non-decreasing to ∞, as x → ∞. I Also, because u(x) need not be continuous, assume there are exponents u0 (x) ≤ u(x) ≤ u1 (x), with ui0 (x) > 0, and u1 (x) = (1+o(1))u0 (x) Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Assumptions to Construct a Lower Bound for md I Write F (x) := P(η ≤ x). Assume F (x) = 1−exp(−u(x)), for u(x) non-decreasing to ∞, as x → ∞. I Also, because u(x) need not be continuous, assume there are exponents u0 (x) ≤ u(x) ≤ u1 (x), with ui0 (x) > 0, and u1 (x) = (1+o(1))u0 (x) I Assume that Fi (x) := 1 − exp(−ui (x)) is concave for large x, each i = 0, 1. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Assumptions to Construct a Lower Bound for md I Write F (x) := P(η ≤ x). Assume F (x) = 1−exp(−u(x)), for u(x) non-decreasing to ∞, as x → ∞. I Also, because u(x) need not be continuous, assume there are exponents u0 (x) ≤ u(x) ≤ u1 (x), with ui0 (x) > 0, and u1 (x) = (1+o(1))u0 (x) I I Assume that Fi (x) := 1 − exp(−ui (x)) is concave for large x, each i = 0, 1. Finally, assume the exponential tail condition lim inf ui0 (x) > 0, i = 0, 1. x→∞ Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Lower bound for md I Define U(u) as the inverse function of u: U := u −1 . Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Lower bound for md I I Define U(u) as the inverse function of u: U := u −1 . Lemma 1: md & U(ln(2d)), as d → ∞. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Lower bound for md I I I Define U(u) as the inverse function of u: U := u −1 . Lemma 1: md & U(ln(2d)), as d → ∞. Proof uses that Fi−1 (t) is convex, i = 0, 1 and “brackets” F −1 (t). In particular, by Jensen’s inequality Z 1 Z 1 −1 2d md = F (t)d(t ) & F1−1 (t)d t 2d /(1 − t02d ) 0 t0 & F1−1 (1 − 1/(2d) + O(1/d 2 )). Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Lower bound for md I I I Define U(u) as the inverse function of u: U := u −1 . Lemma 1: md & U(ln(2d)), as d → ∞. Proof uses that Fi−1 (t) is convex, i = 0, 1 and “brackets” F −1 (t). In particular, by Jensen’s inequality Z 1 Z 1 −1 2d md = F (t)d(t ) & F1−1 (t)d t 2d /(1 − t02d ) 0 t0 & F1−1 (1 − 1/(2d) + O(1/d 2 )). I We obtain, using that U1 := u1−1 grows at most linearly, F1−1 (1−1/(2d)+O(1/d 2 )) = U1 (ln(2d)+o(1)) ∼ U1 (ln(2d)). Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions A Class of Distributions I Now assume u(x) = xv (x) + δ(x), x ≥ x0 [cf. Ben-Ari (2007)], where δ(x) = O(x), and v (x) is twice continuously differentiable and satisfies the following shape, growth, and regularity conditions: v (x), x ≥ x0 % ∞, and xv (x), x ≥ x0 , is strictly convex Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions A Class of Distributions I Now assume u(x) = xv (x) + δ(x), x ≥ x0 [cf. Ben-Ari (2007)], where δ(x) = O(x), and v (x) is twice continuously differentiable and satisfies the following shape, growth, and regularity conditions: v (x), x ≥ x0 % ∞, and xv (x), x ≥ x0 , is strictly convex I growth: lim inf x→∞ xv 0 (x) > 0. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions A Class of Distributions I Now assume u(x) = xv (x) + δ(x), x ≥ x0 [cf. Ben-Ari (2007)], where δ(x) = O(x), and v (x) is twice continuously differentiable and satisfies the following shape, growth, and regularity conditions: v (x), x ≥ x0 % ∞, and xv (x), x ≥ x0 , is strictly convex I growth: I regularity: lim inf x→∞ xv 0 (x) > 0. lim inf x→∞ (xv (x))02 /(xv (x))00 > 1. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions A Class of Distributions I Now assume u(x) = xv (x) + δ(x), x ≥ x0 [cf. Ben-Ari (2007)], where δ(x) = O(x), and v (x) is twice continuously differentiable and satisfies the following shape, growth, and regularity conditions: v (x), x ≥ x0 % ∞, and xv (x), x ≥ x0 , is strictly convex lim inf x→∞ xv 0 (x) > 0. I growth: I regularity: I Theorem 1. Under these conditions, lim inf x→∞ (xv (x))02 /(xv (x))00 > 1. νd ∼ U(ln(2d)), as d → ∞. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Proof of Thm 1 by Log MGF and Proposition 1 I Define a random variable η+ taking values in [x0 , ∞) to have the distribution function F+ (x) := P(η ≤ x|η ≥ x0 ), so λ+ (β) ≥ f+ (β) ≥ νd (η+ )β − ln(2d) ≥ νd (η)β − ln(2d) Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Proof of Thm 1 by Log MGF and Proposition 1 I Define a random variable η+ taking values in [x0 , ∞) to have the distribution function F+ (x) := P(η ≤ x|η ≥ x0 ), so λ+ (β) ≥ f+ (β) ≥ νd (η+ )β − ln(2d) ≥ νd (η)β − ln(2d) I Lemma 2. Define w (x) := (xv (x))0 = xv 0 (x) + v (x). Put U = Ud = U(ln(2d)) for U = the inverse of xv (x), and β = βd := w (U) − M for a constant M > 0.Then λ+ (βd ) . U 2 v 0 (U), as d → ∞. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Proof of Thm 1 by Log MGF and Proposition 1 I Define a random variable η+ taking values in [x0 , ∞) to have the distribution function F+ (x) := P(η ≤ x|η ≥ x0 ), so λ+ (β) ≥ f+ (β) ≥ νd (η+ )β − ln(2d) ≥ νd (η)β − ln(2d) I Lemma 2. Define w (x) := (xv (x))0 = xv 0 (x) + v (x). Put U = Ud = U(ln(2d)) for U = the inverse of xv (x), and β = βd := w (U) − M for a constant M > 0.Then λ+ (βd ) . U 2 v 0 (U), as d → ∞. I Finally, by these two estimates, ln(2d) + U 2 v 0 (U) & w (U)νd , or, u(U) + U 2 v 0 (U) & w (U)νd Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Proof of Thm 1 by Log MGF and Proposition 1 I Define a random variable η+ taking values in [x0 , ∞) to have the distribution function F+ (x) := P(η ≤ x|η ≥ x0 ), so λ+ (β) ≥ f+ (β) ≥ νd (η+ )β − ln(2d) ≥ νd (η)β − ln(2d) I Lemma 2. Define w (x) := (xv (x))0 = xv 0 (x) + v (x). Put U = Ud = U(ln(2d)) for U = the inverse of xv (x), and β = βd := w (U) − M for a constant M > 0.Then λ+ (βd ) . U 2 v 0 (U), as d → ∞. I Finally, by these two estimates, ln(2d) + U 2 v 0 (U) & w (U)νd , or, u(U) + U 2 v 0 (U) & w (U)νd I But u(x) + x 2 v 0 (x) = xw (x), so U & νd . Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Comments on the Proof I Since by the lower bound of Lemma 1, νd & U(ln(2d)), and as a consequence of Proposition 1 and Lemma 2, we have the upper bound νd . U(ln(2d)), the Theorem is proved. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Comments on the Proof I I Since by the lower bound of Lemma 1, νd & U(ln(2d)), and as a consequence of Proposition 1 and Lemma 2, we have the upper bound νd . U(ln(2d)), the Theorem is proved. Comments on the proof of Lemma 2. After integration by parts, we have the formula: Z ∞ exp(λ+ (β)) = C0 β exp(βx − xv (x) + δ(x))dx. x0 R V (2β) R ∞ Split the integral by x0 + V (2β) , where V := v −1 . The second integral is negligible since for x ≥ V (2β), we have x(β − v (x)) + δ(x) ≤ (M − β)x ≤ −(β/2)x . Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Comments on the Proof I The main contribution of exp(λ+ (β)) is thus bounded above by C0 β exp max {x(β + M) − xv (x)} [V (2β) − x0 ] x0 ≤x≤V (2β) Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions Comments on the Proof I The main contribution of exp(λ+ (β)) is thus bounded above by C0 β exp max {x(β + M) − xv (x)} [V (2β) − x0 ] x0 ≤x≤V (2β) I The exponent has a unique maximum by strict convexity of xv (x), and we choose β implicitly by choosing the critical point to satisfy x = xβ = U = U(ln(2d)). This yields w (x) = β + M for w (x) = (xv (x))0 , so the maximum exponent is written U(w (U) − v (U)) = U 2 v 0 (U). Lemma 2 follows after some work to show that under the stated growth condition ln(βV (2β)) = o(U 2 v 0 (U)). Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions The Poisson case I By Stirling’s formula ln(x!) = c + o(1) + (x + 1/2) ln(x) − x, so we find that the Poisson case is represented by v (x) = ln(x) − 1 with δ(x) = O(ln(x)). The growth condition is just satisfied: xv 0 (x) = 1, and the regularity condition (xv (x))02 /(xv (x))00 ∼ x ln(x)2 is easily satisfied. Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage Overview Directed Last Passage Site Model and Directed Polymer Model Relation between f (β) and νd Asymptotic Evaluation of νd for Exponential Weights Asymptotics of νd for a Class of Distributions The Poisson case I By Stirling’s formula ln(x!) = c + o(1) + (x + 1/2) ln(x) − x, so we find that the Poisson case is represented by v (x) = ln(x) − 1 with δ(x) = O(ln(x)). The growth condition is just satisfied: xv 0 (x) = 1, and the regularity condition (xv (x))02 /(xv (x))00 ∼ x ln(x)2 is easily satisfied. I We have U as the inverse of xv (x), or U(u) ∼ u/ ln(u). Therefore νd ∼ ln(2d)/ ln ln(2d). Gregory J. Morrow Department of Mathematics University of Colorado A directed at Colorado polymerSprings approach to the once-oriented last passage