Chapter 6

Renormalization

Previous chapter: Focus on behavior of system close to critical points

; Power laws, critical exponents, relations between critical exponents

Could be described in terms of a scaling hypothesis.

However, the physical basis for the scaling hypothesis was not clear

This section: Theoretical framework that explains origin of scaling laws

; Provides a way to calculate critical exponents

But also: A way of thinking about scale-invariant systems which has
applications far beyond the theory of phase transitions !

6.1 Kadanoff ’s argument for the scaling hypothesis1

Configurations in the 2D Ising model1
T=2T c
upon approaching Tc from above:
Larger and larger clusters ap-
pear, but small clusters are still
present.
T=1.05Tc
; Self-similar structure
; Motivates a description in terms of
successive coarse-graining steps
(Kadanoff, 1966).
T=Tc
; Justifies scaling hypothesis and es-
tablishes a connection to the di-
verging correlation length.
ř ř
Starting point: Ising model βH “ ´ looβJ
moon Si Sj ´ loβH
omoon i Si
xijy
K h

1
Figures in this section are taken from: Kenneth G. Wilson, Scientific American 1979,
158, Problems in Physics with Many Scales of Length

75
76 CHAPTER 6. RENORMALIZATION

Block spins: Combine spins in blocks with side length b.

+++−++ ; ”Coarse-graining”: Every block is characterized
b +−
+−−−++
+−+−−+ by one block spin ř
variable SI “ ˘1.
(e.g., Si “ sign iPI Si )
Example (K. G. Wilson, Scientific American 1979):
T = 1.22 Tc T = Tc T = 0.99 Tc

Block spins: T 1 is higher ... the same ... lower than T

See also https://www.youtube.com/watch?v=MxRddFrEnPc
Assumptions:
(i) ”Blockspins have only pairwise interactions ř with direct řneighbors”
Ñ one can replace partition function tSi u e ´βH by tSI u e´βHl
ř ř
such that tSi u e´βH ř “ tSI u e ř
´βHl

with βHb “ ´Kb xijy SI SJ ´ hb I SI ` const.

Then we have:
‹ Correlation length: ξb “ ξ{b
‹ Functional form of singular part of free energy is the same:
N b´d fs ptb , hb q “ N fs pt, hq, i.e., fs ptb , hb q “ bd fs pt, hq
(ii) Close to the critical point, we postulate: tb “ t byt , hb “ h byh
Reasoning: Simplest relation that ensures tb “ hb “ 0 for t “ h “ 0
and (hb Ñ ´hb , tb Ñ tb ) for h Ñ ´h
Since ξb ă ξ, we must have yt , yh ą 0
( ξb smaller ; phb , tb q further away from critical point)
Note: Just assumptions, no rigorous proof.
In general, only approximately valid
Consequences:
Free energy assumes the form: fs pt, hq “ b´d fs pt byt , h byh q
Now choose: b “ |t|´1{yt ñ fx pt, hq “ |t|d{yt fs p1, h|t|´yh {yt q
; Scaling hypothesis: fs pt, hq “ |t|2´α Ff ph{|t|∆ q
with 2 ´ α “ d{yt , ∆ “ yh {yt , Ff pxq “ fs p1, xq
6.2. BASIC IDEA OF RENORMALIZATION 77

Conclusions: Another motivation of the scaling hypothesis.

Otherwise, not much new insight
– Idea of self similarity (assumption (i)) not new
fl Idea, that diverging correlation length drives singularities (Chap. 5)
– Exponents yt , yh still cannot be calculated.
However: Guides our thinking in a direction that will turn out very fruitful:
Block spin summation Ñ Thinning of degrees of freedom
; Will lead to a better understanding of critical phenomena

6.2 Basic idea of renormalization

Here: First schematic sketch. Systematization of Kadanoff’s idea.
~ “ pK1 , ¨ ¨ ¨ Kn q
Consider a system with degrees of freedom tσu and coupling constants K
Examples
ř ř
– Ising model: βH “ `K1 i Si ` K2 xijy Si Sj ` ¨ ¨ ¨
Degrees of freedom: Si ; Coupling constants: K1 , K2 , ¨ ¨ ¨ (
– Ginzburg Landau model: βH “ dd r g ` 12 p∇Φq2 ` 21 r0 Φ2 ` 14 u0 Φ4 ` ¨ ¨ ¨
ş

Degrees of freedom: Φp~rq; Coupling constants: r0 , u0 , ¨ ¨ ¨

Coupling constants may be zero. However, the set of coupling constants should
be ”complete” in a sense to be defined below.
; Hamiltonian has the form: βH “ const. ` Kα ψα ptσuq
ř

6.2.1 Renormalization group (RG) transformation

Two basic steps
(i) Thinning out: Replace locally bd degrees of freedom by one
~ “
ş d
e.g., tSi uiPI Ñ SI (block spin); Φp~rq Ñ ΦpRq ~rPΩ ~ d r Φp~
rq
R

(ii) Rescale such that the new system with the new degrees of freedom has
locally the same structure than the old one (possibly with new coupling
constants) ; N Ñ N {bd , i.e., V Ñ V {bd
ř ř ´βH ř ´βH 1
ñ New Hamiltonian: βH 1 “ const.1 ` Kα1 Ψα ptσ 1 uq with e “ e
tσu tσ 1 u

Set of coupling constants ”complete” Ñ no new coupling constants

1 N ~1
; Free energy: ( tσu e´βH “ e´βN pg`fs pKq
~ ! 1
´βpN pg`g q` d fs pK qq
“ tσ1 u e´βH “ e
ř ř
b )
~ “ fs pK
ñ fs pKq ~ q d ` g pKq
1 1 1 ~
b
1 ~ Regular contribution from local integration,
with g pKq:
does not contribute to the singular behavior
Together: Defines map: K~ Ñ Rb pKq~
with the properties of a semi-group: Rb1 ˝ Rb2 “ Rb1 ¨b2
(no full group because in general, R cannot be inverted)
78 CHAPTER 6. RENORMALIZATION

6.2.2 Fixed points, RG flow, and critical behavior

~˚
Idea: Critical behavior is associated with fixed points K
of the RG transformation: Rb pK ~ ˚q “ K
~˚ Space of
coupling constants
Fixed point fl Self similar system Real system
In general reached after infinitely many iterations
Fixed point
(Local self similarity is not assumed! )

Consequences for the correlation length ξ

• At the fixed point, one has either ξ ˚ “ 0 or ξ ˚ Ñ 8

R
(Reason: ξ ˚ Ñb ξ ˚ {b “ ξ ˚ only possible for ξ ˚ “ 0, 8)
ξ ˚ “ 0: Trivial fixed point
ξ ˚ “ 8: Critical fixed point
• All points attracted by critical fixed point have ξ Ñ 8
(Reason: ξpKq~ “ bξpRb pKqq~ “ ¨ ¨ ¨ “ bn ξpRn pKqq
~
b
~ ˚ q “ 8, since b ą 1)
“ ¨ ¨ ¨ “ b8 ξpK

Behavior in the vicinity of a critical fixed point

Consider isolated fixed points (generalization is not difficult)

~ “K
Expand about fixed point: K ~ ˚ ` δ K,
~ δK
~ small
ř BRb pKq ~
~ 1 “ Rb pK
RG transformation: K ~ ˚ ` δ Kq
~ “ R b pK
~˚
α BKα δKα
looomooonq ` loooooooomoooooooon
~˚
K ~
“:Mb δ K
~1 “K
ñ K ~ ˚ ` δK ~ 1 “ Mb δ K
~ 1 with δ K ~
linearized RG transformation
Assume Mb can be diagonalized and has real Eigenvalues
(usually correct at critical points. Otherwise, statements below are
not correct)
pνq
Eigenvalue equation: Mb~epνq “ λb ~epνq
pνq pνq pνq
We have: Mbn~epνq “ pλb qn~epνq ñ λbn “ pλb qn
pνq pνq
M1 “ 1 ñ λ1 “ λb0 “ 1
pνq pνq
Together: λbn “ bnyν with yν “ ln λb { ln b
~ “ ř apνq ~epνq
Insert: δ K ν
ñ δK ~ “ ř apνq λpνq~epνq “ ř apνq~epνq byν
~ 1 “ Mb δ K
ν b ν

ñ Some components of δ K~ grow, others shrink Space of coupling constants

relevant
yν ą 0 : growth – ”relevant” directions
yν ă 0 : shrink – ”irrelevant” directions FP
yν “ 0 : ”marginal” directions
irrelevant
6.2. BASIC IDEA OF RENORMALIZATION 79

6.2.3 Relation to scaling hypothesis

Express fs in Eigendirections of M in the vicinity of the fixed point:
~ “ 1d fs pby1 δK1 , by2 δK2 , ¨ ¨ ¨ q “ b´nd fs pbny1 δK1 , bny2 δK2 , ¨ ¨ ¨ q
fs pδ Kq b

Assume we have two relevant scaling fields t and h (e.g., Ising type systems)
ñ fs pt, h, δK3 , ¨ ¨ ¨ q “ b´nd fs pbnyt t, bnyh h, bny3 δK3 , ¨ ¨ ¨ q
Consider t Ñ 0 and choose bn “ t´1{yt (OK since b ą 1)
ñ fs pt, h, δK3 , ¨ ¨ ¨ q “ td{yt fs p1, t´yh {yt h, t´y3 {yt δK3 , ¨ ¨ ¨ q
« td{yt fs p1, t´yh {ht h, 0, ¨ ¨ ¨ q

; Motivates again a scaling hypothesis, however, more systematically than

before. Points at a way to actually calculate critical exponents!

Remark: Here we have made the assumption that one can take the limit
tÑ0
fs p¨ ¨ ¨ , δK3 Ñ 0, ¨ ¨ ¨ q Ñ fs p¨ ¨ ¨ , δK3 “ 0, ¨ ¨ ¨ q without introducing a
new t-dependence. This is not always the case for irrelevant variables.
If fs „ δKxα for certain irrelevant variables, the critical exponents may
change (see Sec. 5.4). Such irrelevant variable are called dangerous.
(Example: In the Ginzburg-Landau theory, u0 is irrelevant at d ą 4,
but it still makes a difference whether u0 “ 0 or u0 ą 0, see Sec. 6.5)

6.2.4 Final Remarks

~ Ñ Rbn pKq
‹ Differential form of the RG equations: Consider K ~ “: Kpl
~ “ bn q
dK~ 1 ~ ´ Rl pKqq ~
ñ RG flow equations: dl “ lim 1 pRlp1`q pKq
l Ñ0
loooooooooooooooomoooooooooooooooon
~
βpKq,independent of l
~
dK ~ with s “ ln l
ñ ds “ βpKq Beta function

‹ Universality:
Many different systems are attracted by the same fixed point. Every fixed
point defines a universality class. In combination with the Ginzburg-
Landau theory, this explains why the universality class (the fixed point),
in general, only depends on the spatial dimension, the symmetry of the
order parameter, and the range of the interactions.
(The assumption that most parameters in the Ginzburg-Landau expan-
sions correspond to irrelevant dimensions can be made plausible by di-
mensional considerations, similar to Sec. 5.4.2, see Sec. 6.5.)

‹ In practice ... it is often not possible to do exact RG transformations. Nu-

merical approximations and/or expansions are necessary. Nevertheless,
the RG approach is valuable because it ...
– Gives physical explanation for the existence of critical exponents,
scaling behavior, and universality
– Provides a starting point for systematic investigations
– Represents a new way of thinking
80 CHAPTER 6. RENORMALIZATION

6.3 Examples of exact RG transformations

Illustration of RG procedure. Exact renormalization is possible, but ”boring”
results. More complex, nontrivial examples will follow later.

6.3.1 One dimensional Ising model

Example of a renormalization group calculation in real space

System: One dimensional Ising chain with periodic boundary conditions

ř ř
βH “ ´K 1N Si Si`1 ´ h i Si “ H̃pK, h; tSi uq pSN `1 “ S1 q
Partition function: Z “ tSi u e´βH
ř

RG transformation

(i) Thinning: Average over spins Si with odd index i

(ii) Rescaling: Sj1 “ S2j
! ř 1 1
ñ Z “ tSi u e´βH tSi u “ tS 1 u e´βH tSj u
ř
j
ř ř (
ř ř ř ř K Si Si`1 `h Si
Hence: S2 ¨¨¨ S3 ¨ ¨ ¨ e “
řS4 hS ř S1 ‰“ ř ‰
“ S2 e 2 S4 ehS4 ¨ ¨ ¨ S1 ph`KpSN `S2 qq S3 ph`KpS1 `S3 qq ¨ ¨ ¨
ř
“ S1 e ‰“ S3 e ‰
“ S2 ehS2 S4 ehS4 ¨ ¨ ¨ 2 coshph`KpSN `S2 qq 2 coshph`KpS2 `S4 qq ¨ ¨ ¨
ř ř

Sj1 “S2j ř h1 Sj1 `K 1 Sj1 Sj`1

ř ř 1
`N g 1 1 1
“ tS 1 u e´H̃pK ,h ,tSj uq`N gpK,hq
ř
“ tS 1 u e
j j
where we choose K 1 , h1 , g such that for all x, y “ ˘1:
h 2g`h1 x`y
xy `K 1
e 2 px`yq
$ 2 coshph ` Kpx ` hqq “: e 2
,
1
’ x “ ´y : 2 coshphq “ e 2g´K piq . /
&
1 1
ñ x“y“1 : eh 2 coshph ` 2Kq “ e2g`h `K piiq
’ 1 1 /
x “ y “ ´1 : e´h 2 coshp´h ` 2Kq “ e2g´h `K piiiq
% -
; Three equations for three unknowns
´H̃pK 1 ,h1 ,tSj1 uq`N gpK,hq
ñ Z “
ř
tSj1 u e

“ 81 ln `24 cosh2 phq coshp2K ` hq coshp2K ´ hq

` ˘
with g
` hq coshp2K ´ hq{ cosh˘2 phq
˘
K 1 “ 41 ln coshp2K
`
1
h1 “ h ` 2 ln coshp2K ` hq{ coshp2K ´ hq
Transformation pK, hq Ñ pK 1 , h1 q “ RpK, hq defines RG flow
Function gpK, hq fl regular contribution due to integration
Needed in order to calculate the free energy βF “ ´ ln Z
(´ βF “ gpK, hq ` 12 gpRpK, hqq ` 14 gpR2 pK, hqq ` ¨ ¨ ¨ “ 1 j
ř8 j
N j“0 p 2 q gpR pK, hqq)

Analysis of RG flow
h
‹ h “ 0 ñ h1 “ 0
K
h ą 0 ñ h1 ą h
h ă 0 ñ h1 ă h
‹ For all h: K 1 ď K
(Check: Set x “ e´4K , y “ e´2h
0 ă x ă 1 ñ 1 ` xy ă 1 ` y, x ` y ă 1 ` y
p1`yq2
ñ x1 “ p1`xyqpx`yq
xąx X)
6.3. EXAMPLES OF EXACT RG TRANSFORMATIONS 81

Fixed points:

‹ K “ 0, h arbitrary (fl T Ñ 8): ξ “ 0

Trivial fixed points, volume phase
‹ K “ 8, h “ 0 (fl T “ 0): ξ Ñ 8 according to Sec. 3.2
Critical fixed point, but unreachable at T ‰ 0

Critical exponents and scaling behavior: Use scaling variables h and x “ e´4K
ñ Critical fixed point at px˚ , h˚ q “ p0, 0q
` ˘ ` ˚ ˘ `δx˘ `δx1 ˘
Linearization: hx “ hx˚ ` δh
`δx˘
ñ δh 1 “ M δh
` ˘ ` yx ˘
with M “ 40 02 “ 02 2yh0 ñ yx “ 2, yh “ 1
ñ Free energy scaling: fs px, hq “ xd{yx Ff ph x´yh {yx q “ x1{2 Ff ph x´1{2 q

6.3.2 The Gaussian model

Example for a renormalization in Fourier space

System: Ginzburg Landau model (Φ4 model) with u0 “ 0

(
βF “ dd r 12 p∇Φq2 ` 12 r0 Φ2 ´ hΦ
ş

Makes sense only for r0 ą 0 (otherwise, Z diverges)

In that case, exactly soluble (”trivial”)
Consider this system in Fourier space, with microscopic cutoff Λ
(
βF “ V1 |~k|ăΛ 12 |Φ~k |2 pr0 ` k 2 q ´ hΦ0
ř

1
ş 1 p2πqd
dd k Ñ
ř
(Prefactor: Discretize p2πqd p2πqd ~ vk
k ~
with v~k “ V
)
1 ř 1
” ś ı V 2
|Φ~k |2 pr0 `k2 q´hΦ0
Partition function: Z “ DrΦse´βF “ 1
ş ş ~
V dΦ~k e k

|~k|ăΛ

RG transformation

(i) Thinning: Integrate over degrees of freedom with short wavelengths

; Calculate partial trace for components Φ~k with Λ{l ă |~k| ă Λ
Easy here, since different ~k modes decouple in βF
ř 1 1 2 2
ñ Z “ 1 1
dΦ~k e V ~k 2 |Φ~k | pr0 `k q´hΦ0
ś ş ś ş (
|~
k|ăΛ{l V
dΦ~k
Λ{lă|~
k|ăΛ V
1 ř 1 |Φ |2 pr `k2 q´hΦ 1 1 2 2 (
“ |~k|ăΛ{l V1 dΦ~k e V |~k|ăΛ{l 2 ~k 0 0 1
dΦ~k e V 2 |Φ~k | pr0 `k q
ś ş ś ş
V
loooooooooooooooooomoooooooooooooooooon
Λ{lă|~
k|ăΛ
π{pr0 `k2 q
1 ř 1 2 2
ś 1
ş ~ |Φ ~ | pr0 `k q´hΦ 0
“ |~k|ăΛ{l V dΦ~k e V |k|ăΛ{l 2 k exppV gpr0 qq
1 ř
` π ˘ 1
şΛ d ` π ˘
with gpr0 q “ V Λ{lă|~k|ăΛ ln r `k2 « p2πqd Λ{l d k ln r `k 2 : regular
0 0

(ii) Rescaling such that new system looks like old system (same cutoff Λ)
~r1 “ ~r{l, ~k 1 “ ~kl, V 1 “ V {ld , Φ1 “ lθ Φ~ 1 (with θ: to be determined)
~k k
1 l´2θ´d ř 1 2 12 2 ´θ 1
l´2θ´d |Φ~ 1 | pr0 `k {l q´hl Φ0
ñ Z “ eV gpr0
ş
k1 |ăΛ
|~
q
ś
|~
k1 |ăΛ V1
dΦ~1 1 e V1 k
k
1
Same structure ñ term 2
p∇Φq2 „ |Φ~k |2 k 2 must look the same
!
ñ l´d´2θ´2 “ 1 ñ θ “ ´p1 ` d{2q
2 ş 1 ř |Φ1 |2 pr0 l2 `k1 2 q´hl1`d{2 Φ10
“ eV gpr0 q |~k1 |ăΛ Vl 1 dΦ~1 1 e V 1 |~k1 |ăΛ ~k1
ś
k

ñ Recursion relations: r “ Rl pr0 q “ r0 l2 , h1 “ h l1`d{2

1
82 CHAPTER 6. RENORMALIZATION

Differential form:
(Rl pr0 q “ r0 l2 ñ dRl pr0 q{dl “ 2r0 l “ 2Rl pr0 q{l)
dr
ñ r0 Ñ rplq with dl “ 2l r i.e., dr{ds “ 2r for s “ ln l
In the same way, one shows: dh{ds “ p1 ` d{2q

Fixed point: r˚ “ 0, h˚ “ 0

Expansion about fixed point:

r0 “ r˚ ` δr ñ δrpsq “ δr e2s “ δr l2
h “ h˚ ` δh ñ δhpsq “ δh ep1`d{2qs “ δh l1`d{2
ñ rp9 tq and h are relevant directions
d
with exponents yr “ yt “ 2, yh “ 1 ` 2

Critical exponents and scaling forms

1
Correlation length: ν “ 1{yt “ 2
cf.6.1
Free energy: fs pt, hq “ l´d fs pt lyt , h lyh q “ |t|d{yt Ff ph|t|´yh {yt q
ñ α “ 2 ´ d{yt “ 2 ´ d2 , ∆ “ yh {yt “ 12 ` d4
β “ 2 ´ α ´ ∆ “ d4 ´ 12 , γ “ 2 ´ α ´ 2β “ 1,
δ “ ∆{β “ d`2
d´2 , η “ 2 ´ γ{ν “ 0

6.4 Renormalization of the 2D Ising model

Example for an approximate RG treatment in real space
(Niemeijer, van Leeuwen, 1974)
ř ř
System: Ising model on the triangular lattice: βH “ ´K xijy Si Sj ´h i Si

RG transformation

(i) Thinning: Combine three spins tS1I , S2I , S3I u

Ñ New triangular lattice
New spin: Majority rule SI1 “ signpS1I ` S2I ` S3I q
?
(ii) Rescaling: Shrink lattice by factor b “ 3
1
ñ Exact RG transformation would read Z “ tS 1 u e´βH “ tSi u e´βH
ř ř
I
” ř ś ı
ñ βH 1 ptSI uq “ ln e´βH δS ,signpř SI q
I α α
tSi u I

Approximate evaluation of RG recursion relations

Separate H into H “ H0 ` V such that

H0 : Contains no ř
couplings between blocks
H0 “ ´K I pS1I S2I ` S1I S3I ` S2I S3I q ´ h Si
ř
V : Rest (H ´ H0 q, treated as ”perturbation”
6.4. RENORMALIZATION OF THE 2D ISING MODEL 83
ř ś
A e´βH0 I δ
tSi u Iq
SI ,signp α Sα
ř
Define xAy0 “ ř ´βH
ś
tSi u e I δS ,signpř S I q
0
I α α
Expectation value with respect to H0 for fixed blockspins SI1
1 N ř 1
Then we have exactly: e´βH “ e 3 A`B I SI xe´βV y0
`A˘ 1
with B “ 2 lnpe3K`3h ` 3e´K´h q ˘ 12 lnpe3K´3h ` 3e´K´h q
(Calculation: Since the
ś individual blocks decouple in H0 , we have
“ I Z0 pK, SI1 q
ř ś
ñ tSi u e´βH0 I δ
SI ,signp α Sα
Iq
ř

with Z0 pK, SI1 q “ S1 ,S2 ,S3 eKpS1 S2 `S1 S3 `S2 S3 q`hpS1 `S2 `S3 q δ
ř
SI ,signpS1 `S2 `S3 q
" 3K`3h *
e ` 3e´K`h : SI1 “ 1
“ ” exppA ` BSI q X)
e3K´3h ` 3e´K´h : SI1 “ ´1

Now approximation: Cumulant expansion xe´βV y0 « e´βxV y0

Then H 1 has the same form as H with new coupling constants:
K 1 “ 2KDpK, hq2
h1 “ BpK, hq ` 12K CpK, hq DpK, hq
`CpK,hq˘ 1 e3K`3h `e´K`h 3K´3h `e´K´h
where DpK,hq “ 2 p e3K`3h `3e´K`h ¯ ee3K´3h `3e ´K´h q

(Calculation: Consider xV y0 in the reference system H0 for given tSI u

ř
– V couples only to neighbor blocks: V “ IJ VIJ
– Averaging: e.g, as in the figure: 1 2
J
βxVIJ y0 “ KxS1I pS2J ` S3J qy 3 1 2
“ KxS1I y0 pxS2J y0 ` xS3J y0 q “ 2KΦpSI1 qΦpSJ1 q I
I y (is independent of α)
with ΦpSI1 q “ xSα 3
0
– One can write ΦpSq as ΦpSq “ C ` DS
(must be possible: Two equations Φp˘1q for two unknowns C, D)
Sα eKpS1 S2 `S1 S3 `S2 S3 q`hpS1 `S2 `S3 q δS 1 , signpS `S `S q
ř
S1,2,3 1 2 3
Iy “
Specifically: xSα I
0 ř KpS1 S2 `S1 S3 `S2 S3 q`hpS1 `S2 `S3 q δ
S1,2,3 e S1 , signpS1 `S2 `S3 q
$ , I
3K`3h ´K`h
e `e
&
e3K`3h `3e´K`h
: SI1 “ 1 .
“ ´e3K´3h ´e´K`h
“: C ` DSI1
%
e3K´3h `3e´K´h
: SI1 “ ´1 -
ñ βVIJ “ 2KpC ` DSI1 qpC ` DSJ1 q “ 2KC 2
` 4KCDpSI ` SJ1 q ` 2KD2 SI1 SJ1
1 N ř 1 ř
ñ βH “ 3 A ` B I SI ´ β IJ VIJ
N
pA ` 6KC 2 q ` looooooomooooooon 2KD2 xIJy SI SJ X )
ř ř
“ looooooooomooooooooon
3
B ` 12KCD I SI1 ` loomoon
N g regular h1 K1

Recursion relations in special cases:

"
` 3K `e´K ˘2 K{2 for K Ñ 0: shrinks
‹ h “ 0: K 1 “ 2 ee3K `e ´3K K «
2K for K Ñ 8: grows
1 2 8
‹ h Ñ 8: K “ 2D K “ 9 K shrinks

ñ Fixed points and RG Flow diagram

h
Trivial fixed points:
ph, Kq “ p0, 0q, p0, 8q, p˘8, 0q K* K

Critical fixed point:

3K ˚ `e´K ˚
ph, Kq “ p0, K ˚ q with p ee3K ˚ `3e ´K ˚
q2 “ 1
2

Analysis of RG results
?
‹ Critical point: Kc “ K ˚ “ ¨ ¨ ¨ “ 14 lnp1 ` 2 2q “ 0.33
1
(Compare exact value: Kc “ 0.27, Bragg-Williams: Kc “ 6 “ 0.17)
84 CHAPTER 6. RENORMALIZATION

‹ Phase diagram: Follows from flow diagram

‹ Critical exponents
` ˘ `K ˘ `K ˚ ˘ ` 1˘
Expand about critical point: δK
δh ˆ“ h ´ h˚ Ñ δK δh1
BK 1 BK 1
˙
`δK 1 ˘ `δK ˘
ñ δh1 “ M δh with M “ BK Bh1
Bh
Bh1
BK Bh pK ˚ ,h˚ q
ˇ ˇ
BK 1 Bh1 ˇ
For symmetry reasons (h˚ “ 0), we have Bh ˚ “ BK
ˇ
ˇ ˇ ˚
h h
ñ M is already diagonal!
Eigenvalues of M:
1
ΛK “ p BK yt
BK qpK ˚ ,h˚ q “ ¨ ¨ ¨ “ 1.62 “: b ñ yt “ 0.882
1
Bh
Λh “ p Bh qpK ˚ ,h˚ q “ ¨ ¨ ¨ “ 3.06 “: by h ñ yh “ 2.034
Compare exact values: yt “ 1, yh “ 15{8 “ 1.875
Remarks
‹ RG treatment allows calculating phase diagrams and critical exponents.
Not exact, but can be improved systematically – e.g. by taking the
cumulant expansion up to second order!
(In that case, RG transformation generates additional interactions
Ñ have to be cut off, further approximations become necessary.
Result is Nevertheless much better: Kc “ 0.26, yt “ 1.04.)
‹ Important: RG transformation must be analytic! (Singularities should
emerge when approaching the fixed point, not prior to that!)
Non-trivial requirement! Among other, it implies that the RG trans-
formation must maintain the symmetries of the system.
(For example, an RG transformation that maps a Heisenberg model
on an Ising model is not allowed!)

6.5 Renormalization of the Φ4 -model

Recall (Φ4 theory and related)
Section 5.4 – Φ4 theory: βF “ dd r t 21 p∇Φq2 ` 12 r0 Φ2 ` 14 u0 Φ4 u
? ?
Rescale: ϕ “ Φ r0 1´d{2 , ~y “ ~r r0
?
ñ βF “ dd y t 12 p∇ϕq2 ` 21 ϕ2 ` 41 ū0 Φ4 u with ū0 “ u0 r0 d´4
At the phase transition, r0 vanishes: r0 Ñ 0
ñ ϕ4 -term diverges for d ă 4, vanishes for d ą 4
(NB: Dimensional analysis: In the latter case (d ą 4), higher order
terms vanish as well. For example, prefactor of ϕ6 scales as r0d´3 .
Section 6.3.2 – Exact renormalization of Gaussian model with u0 “ 0
βF “ dd r t 12 p∇Φq2 ` 21 r0 Φ2 u
Treatment of Φ4 term in Φ4 - theory

a) d ą 4: u0 irrelevant, effectively Gaussian model (”Gaussian fixed

point”). However, u0 is a dangerous irrelevant variable, changes
critical exponents.
6.5. RENORMALIZATION OF THE Φ4 -MODEL 85

b) d ă 4: u0 becomes relevant at the Gaussian fixed point.

; New fixed point (”Wilson-Fisher fixed point”).
Perturbation expansion about u0 and  “ d ´ 4.
Problems:
‚ Expansion diverges ! (already clear from the fact that expansion
must diverge for u0 ă 0! ñ Convergence radius u0 “ 0)
But: First terms of the expansion still improve the results. Be-
yond that, Borel resummation is possible.
‚ Individual terms of perturbation expansion diverge in the limit
Λ Ñ 0 (Λ: microscopic cutoff): ”Ultraviolet divergences”!
; suggests sensitive dependence on Λ
Ways out
– Fourier space renormalization (shown here)
; Degrees of freedom in rΛ, Λ{ls are gradually integrated
out Ñ RG equations. Fixed point is independent of Λ
– Direct elimination of ultraviolet divergences by field theoretic
renormalization (not covered here, detailed treatment, e.g.,
in Binney, Dorwick, Fisher, Newman: The theory of critical
phenomena.

6.5.1 Φ4 -theory in d ą 4 dimensions

Question:

u0 irrelevant Ñ RG group should correspond to that of Gaussian model

(Sec. 6.3.2).
Critical exponents of the Gaussian model:
α “ 2 ´ d2 , β “ d4 ´ 12 , γ “ 1, δ “ d`2
d´2
But: In fact, critical exponents are quite different!
Mean-field exponents, independent of dimension (cf. Sec. 4.3.5)
α “ 0, β “ 21 , γ “ 1, δ “ 3
; How does this fit together?

Answer: u0 is a dangerous irrelevant variable (cf. Sec. 6.2.3).

Scaling form of the free energy in mean-field approximation (according to

dimensional analysis, Sec. 5.3): fs pt, h, u0 q “ l´d fs ptlyt , hlyh , u0 lyu q
with yt “ 2, yh “ 1 ` d2 , yu “ 4 ´ d.
Specifically: Mean-field solution fs “ 12 r0 Φ2 ` 14 u0 Φ4 ´ hΦ
$ a ,
’
’
& r0 {u0 ` h{2|r0 | : r0 ă 0 / /
.
with Φ “’ ph{u0 q1{3 : r0 „ 0 /
’ /
$
%
h{r0 : r0 ą 0 - ,
1 2
a 2 {2r
’
’
’
&
´ 4 r0 {u 0 ´ h r0 {u0 ´ h 0 : r 0 ă 0 /
/
/
.
´1{3
ñ fs “’ ´ 3 u0 h4{3
4 : r 0 „ 0 /
% ´h2 {2r
’ /
: r0 ą 0 -
’ /
0
86 CHAPTER 6. RENORMALIZATION

1 ˆ ?
ñ for u0 ą 0, we can generally write fs pt, h, u0 q „ u0 fs pt, h u0 q
?
´ 14 r02 ´ x r0 ´ x2 {2r0
$ ,
’
’ :tă0 /
/
with fˆs pt, xq “’ ´ 3 x4{3
& .
:t„0 where r0 9 t
4 /
´x2 {2r0
’ /
%
:tą0 -

ñ fs pt, h, u0 q „ l´d fs ptlyt , hlyh , u0 lyu q „ td{yt fs p1, ht´yh {ht , u0 t´yu {yt q
d`yu ? ´ 2yhy`yu
„ t yt 1 fˆs p1, h u0 t
u0
t q
d
Use yt “ 2, yh “ 1 ` 2
, yu “ 4 ´ d
?
„ t2 u10 fˆs p1, h u0 t´3{2 q
„ t2 Ff pht´3{2 q
ˇ
B 2 fs ˇ 0
ñ Specific heat: cH „ Bt ˇh“0 „ t ñ α“0
Order parameter: m „ ´ Bf s
Bh „ t
1{2 F pht´3{2 q
f ñ β “ 1{2
t “ 0 : m „ h1{3 ñ δ“3
xÑ8
(Last line: t Ñ 0, h ‰ 0 ñ m independent of t ñ Ff pxq „ x1{3 ñ m „ h1{3 )

Conclusion:
Since u0 is irrelevant, the RG-calculation still gives the Gaussian fixed
point (Sec. 6.3.2). However, u0 is dangerous and changes the critical expo-
nents. To obtain the correct critical exponents, one needs additional information
(i.e., mean-field solution and Ginzburg criterion!)

6.5.2 Φ4 -theory in d ă 4 dimensions: -expansion

At the Gaussian fixed point, u0 becomes relevant, new RG treatment necessary
(Gaussian fixed point becomes unstable.
Ñ new fixed point: Wilson-Fisher fixed point.)

System: Ginzburg-Landau theory (Φ4 theory) with u0 ą 0

βF “ dd r t 12 p∇Φq2 ` 21 r0 Φ2 ` 41 u0 Φ4 ` ¨ ¨ ¨ ´ hΦu
In Fourier representation ( p2πq
1
dd k “ V1 ~k )
ş ř
d
(
βF “ V1 |~k|ăΛ 12 |Φ~k |2 pr0 ` k 2 q ´ hΦ0
ř
(Gaussian model)
` 41 u0 V14 (Φ4 term)
ř
Φ~k1 Φ~k2 Φ~k3 Φ~k4 V δř4 ~ki ,0
1
~k1 ,~k2 ,~k3 ,~k4
` 16 w0 1
(Φ6 term and higher)
ř
V6
Φ~k1 ¨ ¨ ¨ Φ~k6 V δř6 ~ki ,0 ` ¨ ¨ ¨
1
~k1 ,¨¨¨ ,~k6

Higher order terms (Φ6 etc.) must be included for now, because they
might be generated by RG transformation.

RG transformation (same as 6.3.2)

(i) Thinning: Integrate out Φ~k for Λ{l ă |~k| ă Λ

(ii) Rescale: ~k Ñ ~kl, Φ~k Ñ lθ Φ~k , such that βF retains its form
; term 21 p∇Φq2 remains unchanged.
6.5. RENORMALIZATION OF THE Φ4 -MODEL 87

First (i): Thinning: Integrate out short wavelength modes Φ~k

Separate F “řF0 ` F1 ` V with

βF0 “ V1 |~k|ăΛ{l 12 |Φ~k |2 pr0 ` k 2 q ´ hΦ0
βF1 “ V1 Λ{lď|~k|ăΛ 12 |Φ~k |2 pr0 ` k 2 q
ř

βV “ 14 u0 V14 ~k1 ¨¨¨~k4 Φ~k1 ¨ ¨ ¨ Φ~k4 V δř4 ~ki ,0 ` higher order terms
ř
1
F˜`V gpr0 q ˜
ñ Z “ with e´β F “ e´βV 1 e´βF0
ś 1ş ´β
@ D
V dΦ~k e
|~k|ăΛ{l
“ ś 1
ş ‰
´βF1 “
ř ` π ˘
where V gpr0 q “ ln V dΦ~k e ln r0 `k 2
(see Sec. 6.3.2) Λ{lă|~k|ăΛ Λ{lă|~k|ăΛ
ś ş
@ D ~ dΦ~ e´βF1 A
and A 1 “ śΛ{lă|k|ăΛ ş dΦk e´βF1 ,
Λ{lă|~
k|ăΛ ~
k
´ ¯
β2
@ ´V D 2q ` ¨ ¨ ¨
Cumulant expansion e 1
« exp ´ βxV y1 ` 2 pxV y 1 ´ xV y 1

ñ Expand β F˜ in powers of u0 .
Here: Only very rudimentary sketch of the calculation
For a more systematic treatment involving Feynman diagrams etc.
and better assessment of the approximations below see N. Goldenfeld
Lectures on Phase Transitions and the Renormalization Group
‹ First order: Calculate contribution of Φ4 -term to xV1 y
Separate the contributions to u40 V14 ~k1 ¨¨¨~k4 Φ~k1 ¨ ¨ ¨ Φ~k4 V δř4 ~ki ,0
ř
1
– All |~ki | ą Λ{l (i “ 1, ...4): Can be integrated out completely Ñ regular term
– No |~ki | ă Λ{l: No integration (term remains).
– One or three |~ki | ą Λ{l: Ñ Contribution zero for symmetry reasons (xΦ~k y1 “ 0)
– Two |~ki | ą Λ{l, e.g., |~k1,2 | ă Λ{l, |~k3,4 | ą Λ{l.
ñ Contributes only if q~ :“ ~k3 “ ´~k4 (implying ~k1 “ ´~k2 “: ~k)
(~k3 ‰ ˘~k4 ñ xΦ~k Φ~k y1 “ xΦ~k y1 xΦ~k y1 “ 0,
3 4 3 4
~k3 “ ~k4 not possible, incompatible with ř ~ki “ 0 & |~k1,2 | ă Λ{l)
´ ¯ ř
Single such term contributes u40 V12 qąΛ{l Γpqq V1 kăΛ{l Φ~k Φ´~k
ř

with Γpqq “ x|Φq~ |2 y1 “ ¨ ¨ ¨ “ r V 2 (Gaussian integral)

`4˘ `4˘ 0 `q
In total: 2 “ 6 such terms ( 2 ways to distribute (˘~k, ˘~ q ) on ~k1¨¨¨4 )
Summary: Φ4 term in xVřy1 gives additional contribution to r0
r0 Ñ r̃ “ r0 ` 3u0 V12 qąΛ{l Γpqq (not yet rescaled)
Coefficient u0 remains unchanged in first order of V
‹ Second order: Consider again Φ4 -term
– Contribution of order Opu20 q to r̃ Ñ neglected
(keep only leading contributions of expansion in u0 )
– Contribution to u0 : Leading correction, must be accounted for
... lengthy calculation ...
1
Φ~k ¨ ¨ ¨ Φ~k V δř4 ~k ,0 14 ũ
ř
Ñ V4 1 4 1 i
“ V14 Φ~k ¨ ¨ ¨ Φ~k V δř4 ~k 1
1´9u0 V13
ř ` ř
u
4 0 q,q 1 ąΛ{l ΓpqqΓpq 1 qδ~k ~ q1
1 4 1 i ,0 3 `k4 ´~
q ´~

` ˘
Ñ Approximation: ũ « u0 1 ´ 9u V12 qąΛ{l Γpqq2
ř

‹ Higher order terms in Φ (Φ6 etc.) give higher order contributions to r̃

and ũ (e.g., Φ6 Ñ Opu3 q). Shall be neglected here.
88 CHAPTER 6. RENORMALIZATION

Now (ii): Rescaling – analogous to Sec. 6.3.2:

~k Ñ ~kl, Φ~ Ñ Φ1 “ l´p1`d{2q Φ~ , V Ñ V 1 “ V {ld
k ~ k kl
ñ r1 “ r̃l2 , u1 “ ũl4´d , h1 “ hl1`d{2 (; 1 1 12
V1
rΦ „ l0 , 1
V 13
u1 Φ1 4 „ l0 , h1 Φ1 „ l0 )

Together

Recursion relations - Assume δ :“ l ´ 1 ! 1

r1 “ l2 pr0 ` 3u0 I1 q, u1 “ u0 l4´d p1 ´ 9u0 I2 q, h1 “ hl1`d{2
with I1 “ V12 1 1 δ
ř ş d d´2 C
Γpqq « p2πq d d q r0 `q 2 « 1`r Λ´2 Λ
0
d
Λ{lďqăΛ Λ{lďăΛ
I2 “ V13 1 1 δ
ş
Γpqq2 « p2πq dd q pr0 `q d´4 C
ř
d 2 q2 « p1`r Λ´2 q2 Λ d
0
Λ{lďqăΛ Λ{lďqăΛ
1 21´d
where Cd “ Ωd p2πq d “ πd{2 p d ´1q! (Ωd : Surface of unit sphere)
2
Specifically: C4 “ 1{8π 2
Using l “ 1 ` δ, we can rewrite the recursion relations as
¯
r1 “ r0 ` δ 2r0 ` 3u0 p1`rC0dΛ´2 q Λd´2 ` Opδ 2 q
`
¯
u1 “ u0 ` δ u0 p4 ´ dq ´ 9u0 p1`rC d´4 ` Opδ 2 q
`
d
0Λ
´2 q 2 Λ
` ¯
1 d
h “h`δh 1` 2

Fixed point
Gaussian fixed point: r˚ “ u˚ “ h˚ “ 0
Wilson-Fisher fixed point:
˚ ´2 2
u˚ “ p4 ´ dq p1`r9CΛd q Λ4´d , r˚ “ ´ 32 p1`rC˚dΛ´2 q Λd´2 u˚ , h˚ “ 0.
NB: Approaches r˚ “ u˚ “ 0 for 4 ´ d Ñ 0
; Motivates expansion in  “ 4 ´ d
1 8π 2 1
u˚ “  “ , r˚ “ ´Λ2 , h˚ “ 0
9C4 9 6
Scaling exponents
Linearize again about fixed point and expand
¨ 1
in powers ˛of 
Br Br1
Consider matrix of derivatives ˚ Br01 Bu0 0 ‹
˚ Bu Bu 1
M “˚ ˝ Br Bu 0 ‹‚
‹
0 0
Bh1
0 0 Bh
` Br1 ˘ ` 3u˚ Cd ˘
Br0 FP
“1`δ 2´ p1`r ˚ Λ´2 q2
Λd´4 « 1 ` 2δ ´ δ{3 « p1 ` δq2´{3 “ l2´{3
` Br1 ˘
Bu0 FP
“ 3δ p1`rC˚dΛ´2 q Λd´2 « 3 δ Cd « 0
` Bu1 ˘ ` 18u˚ Cd ˘
d´4 « 1 ´ δ « l´
Bu0 FP
“ 1 ` δ p4 ´ dq ´ p1`r ˚ Λ´2 q Λ
` Bu1 ˘ Cd
Br0 FP
“ 18δ u˚ 2 p1`r˚ Λ´2 q4
Λd´6 « 2
9C4
δ2 «0
` Bh1 ˘
Bh FP
“ l1`d{2 “ l3´{2

ñ yr “ 2 ´ {3, yu “ ´, yh “ 3 ´ {2

; yr , yh ą 0 are relevant scaling fields,
yu ă 0 is irrelevant at the Wilson-Fisher fixed point!
6.5. RENORMALIZATION OF THE Φ4 -MODEL 89

Discussion

‹ Flow diagrams
 ą 0 (d ă 4) r Gaussian fixed point (unstable)

Wilson-Fisher
fixed point

At  “ 0 (d “ 4), Wilson-Fisher fixed point

joins Gaussian fixed point.
 ă 0 (d ą 4) Wilson-Fisher r
fixed point

unphysical
parameter Gaussian fixed point
region
(u < 0)

Wilson-Fisher fixed point becomes unstable and

moves into unphysical parameter region at u ă 0.

‹ Critical exponents at the Wilson-Fisher fixed point, d ă 4

Calculation as before from
yt “ yr “ 2 ´ {3 (since t „ r), yh “ 3 ´ {2
yu ă 0 ñ Scaling field u is irrelevant fixed point
(and also not dangerous)
Results (numbers for 3 dimensions Ñ  “ 1)
α “ 2 ´ d{yt “ {6 ` Op2 q « 0.16
∆ “ yy {yt “ 3{2 ` Op2 q « 1{5
β “ 2 ´ α ´ ∆ “ 1{2 ´ {6 ` Op2 q « 0.33
γ “ 1 ` {6 ` Op2 q « 1.16
δ “ 3 `  ` Op2 q «4
; Much better than Mean-field!
(Correct values in 3D: α “ 0.11, β “ 0.32, γ “ 1.24, δ “ 4.82)