Susceptibility

Linear response of an order parameter to a conjugate field; equivalently a fluctuation or an integrated connected correlation.
Susceptibility

In statistical mechanics, a susceptibility is a linear-response coefficient: it measures how much an of an observable changes when one perturbs the Hamiltonian by a small conjugate field.

Definition (response to a field)

Suppose a field hh couples linearly to an extensive observable MM (e.g. magnetization) through

Hh  =  H0hM, H_h \;=\; H_0 - h\,M,

with equilibrium described by the at inverse temperature β\beta. Let VV denote volume (or number of sites). The susceptibility per unit volume is

χ  =  1VMhh=0. \chi \;=\; \frac{1}{V}\left.\frac{\partial \langle M\rangle}{\partial h}\right|_{h=0}.

This is the prototypical example; more generally, susceptibilities form a matrix of derivatives relating intensive “fields” to extensive “responses,” mirroring the thermodynamic distinction between and variables.

Partition-function and fluctuation formulas

Let Z(β,h)Z(\beta,h) be the for HhH_h. Then

M  =  1βhlogZ(β,h), \langle M\rangle \;=\; \frac{1}{\beta}\,\frac{\partial}{\partial h}\log Z(\beta,h),

so the susceptibility can be written as a second derivative:

χ  =  1βV2h2logZ(β,h)h=0. \chi \;=\; \frac{1}{\beta V}\left.\frac{\partial^2}{\partial h^2}\log Z(\beta,h)\right|_{h=0}.

Using standard identities (see ), one obtains the fluctuation–response relation

χ  =  βV(M2M2)  =  βVVar(M), \chi \;=\; \frac{\beta}{V}\,\Big(\langle M^2\rangle - \langle M\rangle^2\Big) \;=\; \frac{\beta}{V}\,\mathrm{Var}(M),

where the variance is the of MM.

More generally, if hh couples to XX and one measures the response of AA, then

Ah  =  βCov(A,X), \left.\frac{\partial \langle A\rangle}{\partial h}\right| \;=\; \beta\,\mathrm{Cov}(A,X),

expressed via the .

Correlation-function representation

If MM is a spatial sum of local degrees of freedom, for example M=xsxM=\sum_x s_x, then translation invariance gives

1VVar(M)  =  xsxs0c, \frac{1}{V}\,\mathrm{Var}(M) \;=\; \sum_x \langle s_x s_0\rangle_c,

so

χ  =  βxsxs0c. \chi \;=\; \beta \sum_x \langle s_x s_0\rangle_c.

Here sxs0c\langle s_x s_0\rangle_c is the . This identity makes the link to the transparent: if connected correlations decay slowly in space, the spatial sum is large, and the susceptibility becomes large.

Physical interpretation

  • Large susceptibility means the system is easily polarized/ordered by a small field; microscopically, this corresponds to large fluctuations (large Var(M)\mathrm{Var}(M)) and/or long-range correlations.
  • Near continuous phase transitions, the susceptibility often grows strongly or diverges in the , reflecting the growth of correlated domains measured by the correlation length.