Physical Applications: Convexity and Legendre transforms...More Legendre transforms, starting from...

23
Physical Applications: Convexity and Legendre transforms A.C. Maggs CNRS+ESPCI, Paris June 2016

Transcript of Physical Applications: Convexity and Legendre transforms...More Legendre transforms, starting from...

  • Physical Applications: Convexity and Legendretransforms

    A.C. Maggs

    CNRS+ESPCI, Paris

    June 2016

  • Physical applications

    I Landau theories for dielectric response

    I Asymmetric electrolytes with finite volume

    I numerical minimization

  • Exotic dielectric media, water �(q)

    Figure: MD- Bopp, Kornyshev, Sutmann, 1995

    Paradox: is � always positive? See Kirzhnits.

  • Non-local dielectrics

    How to produce �(k)

    I Constraintdiv D = ρ

    I Field energy, (Landau-Ginzburg) for polarization P

    U =(

    E︷ ︸︸ ︷D− P)2

    2+

    P2

    2χ+κp2

    (div P)2

    Water requires κp < 0

    �(k) = 1 +χ0

    1 + k2/k20

  • Variational equations

    A =(Dr − Pr)2

    2+

    Prχ−1r,r′Pr′

    2− φ(div D− ρ)

    Take derivatives:

    δP : − [D− P] + χ−1P = 0δD : + [D− P]︸ ︷︷ ︸

    =E

    +∇φ = 0

    Or

    δP : P(r) =

    ∫χr,r′E(r

    ′)

    δD : E = −∇φ

    Thus in Fourier space

    �(q) = 1 + χ(q)

  • Stability

    A =(Dr − Pr)2

    2+

    Pr χ−1r,r′ Pr′

    2− φ(div D− ρ)

    Consider coefficient of P2:

    UP =Pq(1 + χ−1q )Pq

    2≥ 0

    Grey zone forbidden

    �q = 1 + χq

    Either �(q) > 1 Or �(q) < 0

  • Simplest Landau for negative �

    G =κ

    2P2−α

    2(div P)2+

    β

    2(∇div P)2

    with

    minq

    [1 + (κ− q2α + βq4)

    ]> 0

    giving

    � = 1 +1

    κ− q2α + βq4 0 0.5 1 1.5 2−15−10

    −5

    0

    5

    10

    15

    20

    25

  • Interpretation

    Anti-correlation of hydrogen bond networks?

  • Asymmetric electrolytes

    I Effects of finite volume - ionic liquids

    I lattice models

    I off-lattice formulation

  • Asymmetric Excluded volume in Poisson-Boltzmann

    I Standard entropy function available for symmetric systemsbased on lattice theory

    I Asymmetric generalizations have problematic limitsI Two solutions

    I Recognise that Flory-Huggins entropy has good limitsI Construct theory for an off-lattice equation of state –

    asymmetric Carnahan-Starling

  • More Legendre transforms, starting from general bulk freeenergy

    Full coupled bulk-electrostatic problem:

    F =∫Vd3r (f (c1, c2)− (µ1 − q1ψ)c1 − (µ2 + q2ψ)c2) +

    +

    ∫Vd3r

    (D2

    2ε− ψ∇ ·D

    )Double Legendre transform of free energy gives pressure

    L12[f (c1, c2)]→ −p(µ1, µ2),

    Gibbs-Duhem applied to the Grand potential.Giving generalized PB functional:

    F [ψ] = −∫Vd3r

    (12�(∇ψ)

    2 + p(µ1 − q1ψ, µ2 + q2ψ)).

  • ApplicationsLattice gas, only log divergence at close packingFlory-Huggins -

    βf (φ1, φ2) = φ1 log(φ1)/M + φ2 log(φ2)

    Carnahan-Sterling (single component)

    pV

    NkBT=

    1 + η + η2 − η3

    (1− η)3η = c/c0

    near close packing

    f (c) =c0kBT

    (1− c/c0)2

    Legendre transform applied near close packing:

    −p(ψ) = qc0ψ − γψ2/3

    Compressibility gives very slow, power-law convergence to latticegas expression

  • Asymmetric hard spheres: with Podgornik

    Asymmetric Carnahan-Starling equation of state. Potential or fieldformulation. p(ψ)

    -50 -40 -30 -20 -10 0 10 20-20

    0

    20

    40

    60

    80

    100

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

    L(F

    )(9)

    0

    20

    40

    60

    80

    100

    120

    140

  • Full solution of p(ψ) : local charge density near electrode

    −q = dp(ψ)/dψ ∼ 1− α/ψ1/3, compared with lattice gas

    0 0.2 0.4 0.6 0.8 11=(-eA)1=3

    0

    0.2

    0.4

    0.6

    0.8

    1

    1.2

    (1=Z

    2ec

    0)d

    p=dA

    =!

    q

  • Charged polymer: Numerical optimization

    Model of charged polymer within a thin protein shell - Podgornik

    βf (Ψ, φ) =a2

    6(∇Ψ)2 + vΨ4 − φΨ2 + . . .

  • compare four minimizers

    I

    F1 = (∂f /∂φ)2 + (∂f /∂Ψ)2

    I High order functional in (∇2φ)2

    I Legendre convexification

    I alternating minimization/maximization

  • Numerical results

    N0 100 200 300 400

    L1

    10 -6

    10 -4

    10 -2

    10 0

    10 2

    N10 1 10 2

    t

    10 0

    10 1

    10 2

    10 3

    10 4

    10 5

    Figure: residual errors and stopping time, blue:legendre , green squaredgradient, black alternating nested optimization, red generalized scalarfunctional.

  • Orland-Netz equation

    ∇2Φ− Λe−Ξc(r)/2 sinh Φ = −2ρf (r),

    [∇2 − Λe−Ξc(r)/2 cosh Φ

    ]G (r, r′) = −4πδ(r − r′),

    c(r) = limr→r′

    [G (r, r′)− 1

    |r − r′|

    ],

    implicit matrix equations for a green function

  • Fluctuations in dual Poisson-Boltzmann

    Fluctuation enhanced PB contains interesting physics - imagecharges, self-energies in non-homogeneous dielectric backgrounds

    With Zhenli Xu we have a matrix/iterative solver for theself-consistent problem for Ξ < 5

  • Fluctuations in dual Poisson-Boltzmann

    f = ρeφ− 2c0kT cosh (βqφ)−�

    2(grad φ)2

    One loop correction from: – includes Born energy,

    log det{−div �grad + 2c0q2β cosh qβφ

    }Equivalent form in terms of D

    f =D2

    2�+ L(cosh)[div D− ρ]

    With L(cosh) the Legendre transform of CoshOne loop correction from

    log det

    {1

    �(r)+∇∇L(2)(cosh)

    }Are these related?

  • Determinant identities

    At first sight these determinants seem very different they act onspaces of different dimensions N × N and 3N × 3N

  • Determinant identities

    Compare |−div �(r)grad + c(r)| and∣∣∣ 1�(r) +−grad 1c(r)div ∣∣∣ |c | |�|

    Use |c |∣∣AAT + 1∣∣ and ∣∣1 + ATA∣∣ ∣∣1� ∣∣

    with A = 1√cdiv√� and AT = −

    √�grad 1√

    c

    Proof: by singular values – A = UΣV∣∣1 + AAT ∣∣ = ∣∣1 + UΣΣTU∗∣∣ = ∣∣1 + ΣΣT ∣∣ = ∣∣1 + ΣTΣ∣∣Extra dimensions in matrix have unity on diagonal, samedeterminantSelf consistent one loop? relation between at one loop between φand D.

  • Dynamic casimir

    □ □ □ □ □ □ □ □ □□ □ □ □□ □ □ □ □ □ □ □ □□ □ □ □□ □ □ □ □ □ □ □ □□ □ □ □□ □ □ □ □ □ □ □ □□ □ □ □

    □ □ □ □ □□

    □ □ □ □ □ □ □

    □ □ □ □ □□

    □ □ □ □ □ □ □

    □ □ □ □ □□

    □ □ □□ □ □ □

    □ □ □ □ □□

    □ □ □ □ □ □ □♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢♢ ♢ ♢♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢♢ ♢ ♢♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢♢ ♢ ♢♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢

    ♢ ♢ ♢ ♢

    ♢ ♢ ♢ ♢ ♢ ♢ ♢♢ ♢ ♢

    ♢ ♢ ♢ ♢

    ♢ ♢ ♢ ♢ ♢ ♢ ♢♢ ♢ ♢

    ♢ ♢ ♢ ♢

    ♢ ♢ ♢ ♢ ♢ ♢ ♢♢ ♢

    ♢ ♢ ♢ ♢ ♢

    ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢ ♢○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○○ ○ ○○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○○ ○ ○○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○○ ○ ○○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○○ ○ ○

    ○ ○ ○ ○ ○ ○○ ○

    ○○ ○ ○ ○ ○

    ○ ○ ○ ○ ○ ○○ ○

    ○○ ○ ○ ○ ○

    ○ ○ ○ ○ ○ ○○ ○

    ○○ ○ ○ ○ ○

    ○ ○ ○ ○ ○ ○○ ○

    ○○ ○ ○ ○ ○

    □ N=1000♢ N=2000○ N=4000

    0.01 0.10 1 10 100

    -0.20.00.2

    0.4

    0.6

    0.8

    1.0

    v/Dm

    〈±F(v,L)〉/F C