Entropy in Themodynamics: from Fo-

liation to Categorization
Radoslaw A. Kycia1,2,a

1 Masaryk Univeristy
Department of Mathematics and Statistics
Kotlářská 267/2, 611 37 Brno, The Czech Republic
arXiv:1908.07583v1 [math-ph] 20 Aug 2019

2 Cracow University of Technology

Faculty of Physics, Mathematics and Computer Science
Warszawska 24, Kraków, 31-155, Poland

a [email protected]

Abstract

We overview the notion of entropy in thermodynamics. We start from

the smooth case using differential forms on the manifold, which is the
natural language for thermodynamics. Then the axiomatic definition
of entropy as ordering on set that is induced by adiabatic processes
will be outlined. Finally, the viewpoint of category theory is provided,
which reinterprets the ordering structure as a category of pre-ordered
sets.

Keywords: Entropy; Thermodynamics; Contact structure; Ordering; Posets; Ga-

lois connection

1 Introduction
The notion of entropy (’tropos’ is Greek word transformation) initially appeared
in Thermodynamics to describe the possible direction of the process. In the time,
when this discipline was formed, there was no notion of atoms, and therefore,
the piece of the matter was described using some averaged variables like pressure,
volume, temperature, etc. Currently, we know that these variables come from
the reduction of a large number of degrees of freedom of elements in the piece
of matter (the Avogadro constant NA ∼ 1023 atoms which in classical description
have 3 numbers describing position coordinates, and 3 numbers describing velocity
coordinates) to a few variables mentioned above. The need for pointing out this
’coarse/average’ evolution direction was imminent, and to fulfill this need entropy
was invented, see Fig. 1. Does it mean that if an entropy in theory appears, then

1
Figure 1: System of large degrees of freedoms (atoms) in thermodynamics is
reduced to a few variables.

we are dealing a ’coarse’ (not fundamental) description? We do not know yet.

The history of thermodynamics is full of amazing reasonings that finally lead
to the correct laws of nature. To mention one, Robert Mayer concluded that
heat is energy flow by observing the color of the blood of sailors under different
geographic heights on the ship on which he was a medical doctor. It is also odd
that researchers working in this field end tragically as mentioned Robert Mayer,
Ludwig Boltzmann, or Paul Ehrenfest to mention a few.
Since the pioneering work of Boltzmann that connects thermodynamical en-
tropy with microscopic properties of matter and properties of logarithm function,
this notion appeared to be useful in various disciplines like information theory [22]
or dynamical systems [9]. The increasing importance of this concept is reflected
in bibliometric data of research papers on this subject [24].
This summary will neglect all the classical/physical motivation for thermody-
namics, and we go directly to mathematical concepts associated with the notion of
entropy. We believe that thanks to such approach, we avoid mixing assumptions,
the result of reasoning and ’common knowledge’ in this theory which is common
in physics literature and which leads to the difficulty of grasping these concepts.
Thermodynamics is mature enough to axiomatize fully, and this can be done as
it will be presented below. This presentation is by no means original research -
it is only an overview of the subject and small part of existing literature. Only
the organization of material is perhaps nonstandard and selective. However, it is
hoped it can be treated as a guide for novices (both with mathematical or physical
background) to avoid pitfalls common in this subject.
The overview is organized as follows: First, the geometric meaning of entropy

2
close to original formulation in modern differential geometric terms will be pro-
vided. The presentation will be provided with the context, i.e., geometric struc-
ture of (phenomenological/equilibrium) thermodynamics. Then the axiomatic ap-
proach to entropy will be outlined. Finally, the categorical approach to this subject
will be outlined. In the Appendix, mathematical preliminaries were collected for
the reader convenience, and we advise the reader to look into the Appendix to be
oriented what kind of mathematics is needed to understand the main parts of the
paper.
Ostrava Seminar on Mathematical Physics organized for many years by Diana
Barseghyan, Olga Rossi and Pasha Zusmanovich is a unique platform to exchange
knowledge between mathematicians and physicists. There is also a big audience of
students that motivates speakers to present material in a more pedagogical way,
including also the context of the research. One of my talk, which was a pleasure
to deliver, was about entropy and Landauer’s principle. This overview paper can
be treated as a basic introduction and guide to the subject.

2 Smooth case
We start from describing Thermodynamics in the natural setup of smooth dif-
ferential manifolds with contact structure on it. In these terms, although not so
precisely due to lack of mathematical language development, the fathers of ther-
modynamics were thinking. The presentation follows closely [18, 7, 4, 2, 3, 8].

2.1 Space
In thermodynamics, we identify some system from the environment by marking
more or less formal borders with some specific physical properties (e.g., heat con-
tact or permeability of particles). Such a system should be macroscopically uniform
in the sense of its physical and chemical properties - so-called simple system. By
distinguishing such a system, we can describe it by some variables depending on
the physical context. The common feature of these variables is their uniform be-
havior under the scaling group R+ , which reflects the physical property that the
scaling of the system scales its internal parameters, e.g., scaling the system scales
its volume or energy. These variables are called extensive. Call these variables
{Xi }ni=0 . If there is another extensive variable X, then it must depend on the
previous X = X(X0 , . . . , Xn ) and scales as X(λX0 , . . . , λXn ) = λX(X0 , . . . , Xn )
where λ ∈ R+ that it should be truly extensive.
The common choice of extensive variables are

• U - energy of the system;

• V - volume;

• N - number of particles;

3
 The first assumption is that
Assumption 1. In equilibrium state, the system is fully described by some set of
extensive variables.
Where equilibrium state is attained when the system is left on its own and
relaxes attaining this state without any further change of extensive variables. The
direction in which the evolution of the system left on its own is described by
entropy introduced later.
If a system is composed of more simple systems then the number of variables
is a common set of all the variables, e.g., we have sum of all energies as the total
energy of the full system and volumes of each subsystem.
The system (simple or compound) can interact with the environment by ex-
changing energy. One ’directed’ way of transferring energy is work made by the
environment on the system. These works are described by work 1-forms that relate
change of extensive parameters of the system with work done on the system. In
local coordinates:
W = Pi dX i . (1)
The coefficients Pi are called intensive variables and describe ’generalized forces’
of environment that act on the system. They do not scale. Note that it is not
necessary that W is an exact form and therefore work may depend on the path
along which it is integrated.
The common choice of intensive variables are
• p - pressure; associated with change of volume V ;
• −µ - chemical potential describing density of work done by changing the
number of particles in the system by adding/removing particles/elements
from/to environment or modified by chemical reactions; associated with the
number of particles N ;
The second way of energy transfer between system and environment is the
heat transfer described by a 1-form Q. We will see below that Q can be written
in terms of work form: Q = T dS, where T is the absolute temperature (intensive
variable) and S is the entropy (extensive variable). In physical terms, Q is the
transfer of kinetic energy at the level of atoms/molecules. The detailed description
of this transfer in thermodynamics is neglected by reducing microscopic degrees
of freedom to a few macroscopic ones. Therefore some additional law has to be
introduced that control such transfer. It is done in terms of entropy and the Second
Law of Thermodynamics.
In thermodynamics, the system is described by energy U and 2n pairs of asso-
ciated intensive-extensive variables. These are local coordinates on 2n + 1 dimen-
sional manifold. Each point describes the system. From physics, it is assumed
Assumption 2. The equilibrium state is described as a point in 2n+1 dimensional
smooth manifold M called the space of states.

4
 Local coordinates are usually taken to be (U, (T, S), (p, V ), (µ, N ), . . .), where
intensive-extensive pairs were grouped.
In order to compare systems in equilibrium, we introduce the Zeroth Law of
Thermodynamics
Axiom 1. If a system A is with a thermal equilibrium with B and B with C, then
A is in thermal equilibrium with C.
The thermal equilibrium of two systems means that there is no heat flow Q
between systems that are connected by thermal conducting material. The Law
means that the relation of ’being in thermal equilibrium’ is transitive. It is also
obviously reflexive and symmetric, and therefore is an equivalence relation. It
allows us to define tools called thermometers that measure empirical temperature,
which represent exactly these equivalence classes. This empirical temperature will
be related to the (absolute) temperature T below.

2.2 Processes
The next step is to consider the change in the system that is described by paths
in the space of states called thermodynamical processes.
(Equilibrium) thermodynamics is only occupied with quasi-statics processes
which can be represented by curves in the space of states. In physical terms, they
can be considered as physical/chemical processes which occurs ’slow enough’ that
in every step the system and environment are in equilibrium or relax ’fast enough’
to equilibrium in every step of the process. It is only an idealization. On the
contrary, non-quasi-static processes cannot be described as a path in the space
of states. They can only be marked as initial and final points if these points are
equilibrium states. This peculiarity is connected with the fact that the points in
the space of states describe only equilibrium states.
The other distinction is according to reversibility. The process is:
• reversible process - if it can be conducted in both directions when all variables
(intensive of system and extensive of environment) can be returned to initial
values in local description;

• irreversible process - it cannot be reversed;

For quasi-static processes we have a curve γ in the space of states M that we
assume to be piecewise smooth which is usual assumption to calculate:
R
• ∆Q(γ) := γ Q - total heat transfer in the process;
R
• ∆W (γ) := γ W - total work done in the process;
Note that these definition are not valid when there is no curve along which 1-forms
Q and W can be integrated, i.e. for non-quasi-static processes.
Some examples of thermodynamic processes are as follows [7]:

5
 • Quasi-static adiabatic process - in this case no heat is exchanged, that is
∆Q(γ) = 0;

• Heating at constant volume - it is quasi-static process which

R for the case of
simple system without exchange of particles ∆W (γ) = γ pdV = 0;

• Non-quasi-static process - no path in M therefore no ∆W and no ∆W can

be calculated. Only the the difference of energy between initial and final
state of the process can be defined.

In technical applications, the important are closed paths that are called ther-
modynamical cycles and describes the cyclic work of engines. They are also crucial
in the formulation of the Second Law of Thermodynamics below.

2.3 The First Law of Thermodynamics

The first fundamental law of thermodynamics describes from the physical point of
view the conservation of energy during the quasi-static process, namely

Axiom 2. (First Law of Thermodynamics)

dU = Q − W (2)

We stated this law as an Axiom since on the physics side it is a fundamental

law of nature, and on the mathematical side it is the unquestionable statement.
For quasi static processes described by the curve γ in M with the initial point
x and the final point y integrated version of (2) is

∆U (γ) := U (y) − U (x) = ∆Q(γ) − ∆W (γ). (3)

This is due to the fact that dU is exact form and therefore its integral depends on
endpoints of the curve.
In expanded form (2) can be written in local coordinates as

dU = Q − pdV − µdN. (4)

In this context we can reinterpret the properties of the processes:

• Quasi-static adiabatic process - ∆Q(γ) = 0 and therefore ∆U = −∆W ;

R
• Heating at constant volume - ∆W (γ) = γ pdV = 0 and therefore ∆U = ∆Q;

Note that the quasi-static adiabatic process converts all total energy of the system
to the work that can be extracted to the system. The restrictions on this process
prevents to construct ’perptuum mobile’ and is controlled by the Second Law of
Thermodynamics described below.

6
 We now turn to finish mathematical description of state space. On 2n + 1
dimensional space M have the form

θ := dU − Q + W. (5)

The volume form in M can be given by

θ ∧ (dθ)n 6= 0. (6)

Therefore θ defines contact structure on M or equivalently J 1 (N ), where N has

local coordinates as extensive variables (U, V, N ). This leads to the final definition
of the space of states for thermodynamics

Definition 1. The state of space in thermodynamics is described by even dimen-

sional space M with contact form θ that fulfills θ ∧ (dθ)n 6= 0.

We can now reconstruct the conservation law of the First Law of Thermody-
namics, namely, using Darboux theorem for contact manifolds (see Appendix) we
have that there are local coordinates (X0 , (X1 , P1 ), . . . , (Xn , Pn ) that the canonical
form of θ is
X n
θ = dX0 − Pi dX i . (7)
i=1

Comparing with (5) we have that X0 = U etc.

In this space the submanifold Φ describing physical system in equilibrium ful-
fills
Φ∗ θ = 0, (8)
that is, physical systems are described by such submanifolds of M that preserves
energy/The First Law of Thermodynamics. In the case of a non-degenerate ther-
modynamical system, it is assumed that

Assumption 3. Non-degenerate thermodynamical system is described by maximal

dimension subspace in the contact space of states of dimension 2n+1, i.e., Legendre
submanifolds of dimension n.

The Legendre submanifold is defined by providing X0 = X0 (X1 , . . . , Xn ) or

using (7) we get equations of state
 ∂X
 P1 = P1 (X1 , . . . , Xn ) = ∂X01
... (9)
∂X0
Pn = Pn (X1 , . . . , Xn ) = ∂Xn .


This can be viewed as the equivalence of holonomic sections of jet space and
Legendre submanifolds on contact space - see Appendix.
The last remaining issue is the direction of heat transfer, which is resolved by
the Second Law of Thermodynamics outlined in the next subsection.

7
2.4 The Second Law of Thermodynamics
We now put some restrictions on quasi-static adiabatic paths/processes γ that
are described by γ ∗ Q = 0. All tangent vectors to such paths are in Ker(Q) and
define some distribution on M . Since adiabatic processes along any paths are not
possible therefore Caratheododry formulated the following version of the Second
Law of Thermodynamics
Axiom 3. [7] (Second Law of Thermodynamics)
In a neighborhood of any state x ∈ M there is state y that is not accessible from x
via quasi-static adiabatic paths γ that γ ∗ Q = 0.
Using the Caratheodory’s theorem on accessibility, we get that the distribution
Ker(Q) is integrable (defines holonomic constraints in M ) or, put another way,
Q ∧ dQ = 0. (10)
This is also equivalent to the statement that
Q = T dS, (11)
where T is an integrating factor (nonsingular function on M ) called the absolute
temperature, and S is called the entropy. It means that S = const defines local
leaf of the distribution on which quasi-static adiabatic paths lie.
Considering two simple systems with thermal contact (no adiabatic border). It
can be shown that are in equilibrium if their absolute temperatures T are the same
[2]. This defines equivalence classes as in the Zeroth Law of Thermodynamics, and
therefore absolute temperature can be used as empirical temperature.
There is a stronger version of this law by Kelvin that implies [7] Caratheododry’s
version, namely,
Axiom 4. In no quasi-static cyclic process can a quantity of heat be converted
entirely into its mechanical equivalent of work.
It can be shown that the foliation exists globally and is not pathological by
taking coordinates that consist quasi-static adiabatic paths lying in S = const
leaves and transverse to them paths of heating at a constant volume that change
Q and therefore are not adiabatic, see [7] for details. This shows that entropy S
and T are globally defined on M .
The important conclusion that will be a link between classical entropy and its
axiomatic definition in the next section is
Theorem 1. [7] If a state y results from x by any adiabatic process (quasi-static
or not), then S(y) ≥ S(x).
We, therefore, have that in an isolated (i.e., adiabatic) system entropy cannot
decrease when achieving equilibrium, which is commonly known the version of the
Second Law of Thermodynamics. Here it is presented as a conclusion from a more
geometric formulation of this law.

8
2.5 Symmetries and thermodynamic potentials
Having defined the fundamentals of thermodynamics, we can provide some exam-
ples of different choices of variables that do not change thermodynamics. They are
useful if we use different variables to observe the system. They should not change
the contact structure, and therefore, they are contact symmetries mentioned in
the Appendix.
The most useful is the Legendre transformation that interchange the role in
extensive-intensive pair of variables. This transformation also modifies X0 = U
variable giving a new, so called, thermodynamic potential. We present a few exam-
ples in case of constant number of particles (system boundaries are not permeable
- µ = 0) for simplicity:

• The transformation p ↔ V gives thermodynamic potential called the En-

thalpy X̃0 = H := U + pV and the contact form θ̃ = dH − T dS − V dp. It
is useful to observe the system on the submanifold p = const.

• The transformation T ↔ S gives thermodynamic potential called the Helmholtz

potential/Free energy X̃0 = F := U − T S and the contact form θ̃ =
dF + SdT + pdV . It is useful to observe the system on the submanifold
V = const.

• The transformation p ↔ V and T ↔ S gives potential called the Gibbs

potential X̃0 = G := U +pV −T S and the contact form θ̃ = dG+SdT −V dp.
It is useful to observe the system on the submanifold T = const and p =
const.

2.6 Examples
The thermodynamic relations result from taking exterior derivative of the con-
tact form pulled-back on the Legendre manifold that describes thermodynamical
system. As an example consider the standard contact form

θ = dU − T dS + pdV. (12)

The Legendre manifold Φ is given by equations of state T = T (S, V ) and p =

p(S, V ). Then since Φ∗ θ = 0 and Φ∗ dθ = dΦ∗ θ = 0 we get
 
∗ ∂T ∂p
Φ dθ = − dS ∧ dV = 0, (13)
∂V ∂S

which gives one of the Maxwell relation

∂T ∂p
=− . (14)
∂V ∂S

9
Since Φ is also given by U = U (S, V ) from (12) we get
   
∗ ∂U ∂U
Φ θ= − T dS + + p dV = 0, (15)
∂S ∂V
∂U ∂U
which means that T = ∂S and p = − ∂V . Then (14) can be written as

∂2U ∂2U
= , (16)
∂S∂V ∂V ∂S
which is a tautology for smooth U . In general Maxwell realations can be used as a
consistency check of equations of motion - if they define the Legendre submanifold.
Another example is the ideal gas which has the equation of state

pV = N RT, (17)

where N is the number of moles of the gas, and R is the universal gas constant.
This is not enough for the definition of the Legendre submanifold, and another
relation is provided
3
U = N RT. (18)
2
These equations are provided for the Lagrange manifolds given by S = S(U, V ),
which gives
∂S 1 3N R ∂S p NR
= = , = = . (19)
∂U T 2U ∂V T V
One can easily check that the mixed second derivatives agree.
For more examples, one can look e.g. into [18, 2] or for more physical view [4].

3 Axiomatic approach
The above description of entropy can be axiomatized. Our presentation in this
section closely follows [16] and [17].
We start from the definition of a simple system as in the previous section. The
states of such system are points X, Y, Z, . . . inside the space of states Γ. Then we
fix on the set Γ the structure of the space R2n+1 , where one variable is the energy
U and the remaining 2n variables are extensive-intensive pairs.
On such space we can introduce scaling by λ, µ ∈ R+ that is multiplication
group action Γ1 = Γ, (Γλ )µ = Γλµ . The scaled state λX consists of all extensive
variables scaled and all intensive variables unaffected. Two systems Γ1 and Γ2
can be composed, and then the composed system is described by points from the
Cartesian product Γ1 × Γ2 .
The fundamental notion needed for the definition of entropy in [16, 17] is
adiabatic accessibility
Definition 2. State Y is adiabatically accessible from X if the only result of the
transition is a work done. We denote it X ≺ Y .

10
 This definition does not involve heat since it was not defined yet. Besides, the
relation ≺ is intended to be some ’ordering’ to be specified later.
We can further define

Definition 3. • Irreversible adiabatic process: X ≺≺ Y if X ≺ Y and not

Y ≺ X;

• Adiabatic equivalence: X ∼ Y if X ≺ Y and Y ≺ X;

In order to introduce entropy S : Γ → R the relation ≺ is assumed to fulfill

the axioms [16, 17]:
• Monotonicity: X ∼ X

• Transitivity: X ≺ Y and Y ≺ Z then X ≺ Z

• Consistency: X ≺ X 0 and Y ≺ Y 0 implies (X, Y ) ≺ (X 0 , Y 0 )

• Scaling invariance: λ > 0 and X ≺ Y implies λX ≺ λY

• Splitting recombination: X ∼ (λX, (1 − λ)X)

• Stability: if (X, Z) ≺ (Y, Z 0 ) then X ≺ Y for  → 0+ .

Up to now, the relation ≺ is a partial order, however it can be made a total
ordering by the following Comparison ’Hypothesis’ that can be proved using the
definition of a simple systems and the Zeroth Law of Thermodynamics [17]

Theorem 2. We say that the Comparison Hypothesis (CH) holds for a state-space
Γ if all pairs of states in Γ are comparable.

These assumptions/hypothesis implies the existence of entropy

Theorem 3. [17, 16] A function S : Γ → R called entropy exists under assumption

of the above axioms and Comparison Hypothesis, and fulfills:
• Monotonicity: X ≺ Y ⇔ S(X) ≤ S(Y )

• Additivity: S(X, Y ) = S(X) + S(Y )

• Extensibility: S(λX) = λS(X)

The last issue is to agree all local entropies and check if the global entropy can
be defined. This done in the following

Theorem 4. [17, 16] Assume that CH holds for all compound systems. For each
system Γ let SΓ be some definite entropy function on Γ. Then there are constants
aΓ and B(Γ) such that the function S, defined for all states of all systems by affine
transformation
S(X) = aΓ SΓ (X) + B(Γ), (20)

11
for X ∈ Γ , satisfies additivity (2), extensivity (3), and monotonicity (1) in the
sense that whenever X and Y are in the same state-space, then

X≺Y ⇔ S(X) ≤ S(Y ). (21)

Total ordering ≺ of adiabatic processes and the existence of entropy S that

fulfills monotonicity (for simple systems) is a link with Theorem 1 of smooth case.
It is also a starting point to define entropy in terms of category theory, which will
be the subject of the next section.

4 Categorification
In this section, we review some concepts from [13]. For background from category
theory see [23] or [20].
We will consider only a simple (i.e., not compound) systems for simplicity.
This approach is based on the definition of Poset (pre-ordered set) as a category:

Definition 4. Poset (pre-ordered set) (P, ≺) is a set P with order relation ≺. The
arrow x → y fo rx, y ∈ P is present when x ≺ y.

We will use the definition for ≺ being a total order since this is the case for
entropy from previous sections. Then the ordering relation/the arrow x → y is
defined only when y is adiabatically accessible from x.
If scaling of the system is considered then instead of Poset, the G-Poset cate-
gory has to be considered [1]. The first step is to define a set with group action -
a G-Set [5]

Definition 5. System space is the object of G-Set category, i.e., {Γ, (R+ , ·, 1)},
where the multiplicative group acts on state-space Γ.

Under the assumption from previous section on the poset the ordering is in-
duced by the entropy S : P → R, and therefore we can define

Definition 6. [13] The entropy system is the object of G-Pos category, which
objects are G = (Γ, 4), with preserving ordering group (R+ , ·, 1) action1 , where the
(partial or) total order is given by the entropy function S : Γ → R.

Hereafter we restrict ourself only to Posets for simplicity. For the general case
of G-Posets see [13].
Up to now this is only rephrasing of previous section in terms of ’abstract non-
sense’, and it does not introduce anything new. The situation, however, changes
when we consider more than one entropy system. In this case, we have two or
more posets that can represent differently (and not necessary originating from
1
If for X, Y ∈ Γ there is X 4 Y , then for λ ∈ R+ there is λX 4 λY .

12
thermodynamic) entropy systems. We can ask what is minimal mapping (Func-
tors of these Posets) that preserve ordering, and therefore entropies that introduce
these orderings. It occurs that minimal ’relation’ that preserves these orderings in
both directions is the Galois connection [23, 21]. It can be seen as a basic exam-
ple/a ’prototype’ of adjoint functors. The Galois connection rewritten in terms of
orderings induced by entropy functions has the following form
Definition 7. [13] The Landauer’s connection and Landauer’s functor
Entropy system G1 = (Γ1 , S1 ) is implemented/realized/simulated in the entropy
system G2 = (Γ2 , S2 ) when there is a Galois connection between them, namely,
there is a functor F : G1 → G2 and a functor G : G2 → G1 such that F a G.
In terms of the entropy it is given as
S2 (F c) ≤ S2 (d) ⇔ S1 (c) ≤ S1 (Gd). (22)
We name the functors F and G the Landauer’s functors.
The Galois connection usually appears in logical/model theory considerations
when we have a Poset of some axioms, and we implement them on a Poset of models
that realize these axioms [23]. The ordering is then provided by the ’strength’ of
axiom and model. In this vein, we can use the Landauer’s connection to relate
some abstract entropy model with its implementation on the physical system with
entropy. If such a connection between these two levels model-realization exists,
then the change in entropy at the level of the model is transferred through the
Ladauer’s connection to the change in entropy in the physical realization level. This
was the original idea of Landauer [15, 14], who deduced that any irreversible logical
operation at the level of Shanon-entropic system generates a physical heat. In
terms of the Landauer’s connection this heat is generated by the change of entropy
in physical part of the device that implements this logical system. Therefore, the
categorical approach makes a sharp distinction, in which part of the compound
entropic system such Landauer’s heat is generated. This result also explained
Maxwell’s demon paradox [13].
This sketch presents only one application of the connection. More details and
examples from physics, computer science, and biology can be found in [13].

5 Summary
In this paper, we presented the road from entropy in terms of thermodynamics
to its categorification. We started from the foundations of thermodynamics and
entropy that relies on contact geometric structure. Having understood the mo-
tivation, the axiomatic approach to entropy was presented, which emphasize the
ordering of equilibrium states by the adiabatic processes. Finally, this ordering was
used to reformulate the system with entropy in terms of pre-ordered sets - Posets.
Two such Posets can be Galois connected by functors that preserve orderings, and
therefore entropies. This connection can be used in various interesting contexts.

13
Acknowledgments
I would like to thanks Valentin Lychagin for pointing me in this interesting subject
and fruitful discussions. This overview was written thanks to the encouragement
of Pasha Zusmanovich and warm positive feedback of Lino Feliciano Reséndis
Ocampo. This research was supported by the GACR grant GA19-06357S and
Masaryk University grant MUNI/A/1186/2018. I also thanks to the PHAROS
COST Action (CA16214) for partial support.

A Differential forms
The mathematical structure of equilibrium thermodynamics is the theory of differ-
ential forms on contact space and its integrability. This section outline the theory,
and interested reader is referred to various sources, including [6]. All theorems
are local, which is convenient for applications. Therefore Reader can assume that
every manifold is reduced to its open subset or equivalently (by diffeomorphism)
every manifold is an open subset of Euclidean space. Therefore N will have (local)
coordinate chart (x1 , . . . , xdim(N ) ).

A.1 Frobenius theorem

The basic problem in exterior calculus is to check complete integrability of an
exterior system {ω1 , . . . , ωn }, that is the existence of a submanifold given locally
by n relations Φ := {gi (x) = ci , i = 1 . . . n}, for constants ci , on which the exterior
system vanishes Φ∗ ωi = 0 for i = 1 . . . n. This is given by
Theorem 5. [6] Exterior system {ω1 , . . . , ωn } is completely
P integrable iff there
exists a nonsingular matrix Aij of 0-forms that ωi = j Ai,j dgj .
For a system given by 1-form Q complete integrability means that there exists
an integrating factor (nonsingular 0-form) T such that Q = T dS. This fact will
be useful in defining entropy.
This can be reformulated in terms of differential ideal. We say that the set I
is the differential
P ideal defined by the set of 1-forms {ω1 , . . . , ωn } if for η ∈ I we
have η = i Ai ωi for 0-forms Ai . In this terms the Frobenius theorem controls
complete integrability of the differential ideal defined by the exterior differential
system, namely,
Theorem 6. [6] The ideal I is integrable iff dI ⊂ I.
This means that the ideal I isP closed under exterior derivative, i.e., dη ∈ I if
η ∈ I. This also means that dωi = j Aij ωj for 0-forms Aij or dωi ∧ω1 ∧. . .∧ωn = 0
for i = 1 . . . n.
An alternative version of the Frobenius theorem is formulated for distributions.
Define a vector space D = Span(Ker(ω1 ), . . . , Ker(ωn )). This means that at each

14
point of the space we define a vector subspaces and we are asking if these subspaces
are tangent to some manifold that is an integrable manifold of the distribution D.
Then the Frobenius theorem has the form

Theorem 7. [6] The distribution D is integrable iff [D, D] ⊂ D.

This means that taking all possible vectors from the distribution (which can
be associated with infinitesimal transformations on the manifold), by making its
bracket we cannot get a new vector (infinitesimal transformation) that is out-
side the distribution D. This observation gives the Caratheododry’s theorem on
accessibility states

Theorem 8. [7, 6] If in the neighborhood of any point there are points not acces-
sible by paths which have tangent vectors in the distribution, then the 1-form θ is
integrable (θ ∧ dθ = 0).

Summing up, if the distribution/exterior differential ideal is integrable, then it

defines foliation of the manifold/holonomic constraint. If we intend to stay on the
leaf of the foliation then we cannot reach all points in the neighborhood of the leaf
since then we would leave the leaf moving with tangent vector that not lie in the
distribution. However, this statement is local. Foliation can consist of one leaf as
in the case of the curve with an irrational tangent coefficient that wraps densely on
the torus - this defines initial submanifold [10]. For defining the global structure of
the leaves, and to assure that they are proper submanifolds some additional work
must be done.
The Frobenius theorem will be useful in proving the existence of entropy, which
is the Second Law of Thermodynamics.

A.2 Darboux theorem

The next important theorem is the Darboux theorem that describes local canon-
ical form of the differential 1-form defining contact and symplectic structures on
manifold. We present only version for contact form

Theorem 9. [6]
For a 1-form ω fulfilling ω ∧ (dω)n 6= 0 and (dω)n+1 = 0 there exists n + 1
local functions {Xi (x)}ni=0 and n functions {Pi (x)}ni=1 on the manifold M with
coordinates xi such that the form ω has representation
n
X
ω = dX0 + Pi dXi . (23)
i=1

These functions can be used to introduce new coordinates on the manifold in

which ω has simper form.

15
A.3 Contact structure
We define

Definition 8. The pair (M, θ) where M is even dimensional manifold of dimen-

sion 2n + 1 and θ is non-degenerate 1-form that fulfills θ ∧ (dθ)n 6= 0 is called
contact manifold.

Since dim(M ) = 2n + 1 and deg(θ ∧ (dθ)n ) = 2n + 1 therefore (dθ)n+1 = 0. We

can use the Darboux
P theorem to conclude that locally we can introduce coordinates
that θ = dX0 + ni=1 Pi dXi .
The contact space and the contact form are in Thermodynamics defined by
the First Law of Thermodynamics.
The contact structure is solvable by a submanifold Φ that fulfill Φ∗ θ = 0. We
can ask about the maximal dimension of such submanifold. This is controlled by
the following:

Theorem 10. [11] Maximal submanifold in 2n + 1 dimensional contact manifold

M has dimension n and is called Legendre submanifold.

A.4 Contact structure vs Jet space

We finish this overview of differential geometry by discussing the rudiments of jet
spaces. This presentation is mainly based on [12, 11].
Consider n dimensional manifold N and smooth functions on the manifold
C ∞ (N ). In local coordinates (x1 , . . . xn ) define the ideal
( )
k ∞ ∂ |σ| f
µa := f ∈ C (N )| σ (a) = 0, 0 < |σ| < k , (24)
∂x

|σ| |σ| f
where multiindex σ = (σ1 , . . . , σn ), |σ| = ni=1 σn and ∂∂xσf := ∂(x1 )σ∂1 ...(x
P
n )σn .

Now define the k-th jet of functions at x = a as the quotient

Jak (N ) := C ∞ (N )µk+1
a . (25)

The equivalence classes [f ]ka ∈ Jak (N ) represent the functions that have the same
derivatives/contact at x = a up to order k, in other words, they Taylor series at
x = a agrees up to order (x − a)k . For example for dim(N ) = 1, [x]i0 = [sin(x)]i0
for i = 0, 1, 2 but disagree for i = 3.
The k-jet of function on N is defined as
[
J k (N ) = Jak (N ). (26)
a∈N

It is a fiber bundle π : J k (N ) → N with the obvious projection.

16
 We can now describe J k (N ) locally by taking derivatives as coordinates and
defining the ideal of 1-forms (the Cartan distribution) that ’sets’ these coordinates
to derivatives. For simplicity consider J 1 (N ). The local coordinates are (xi , y, yi )
∂
where the new coordinates pi are associates with derivatives ∂x i by the Cartan
distribution
ω = dy − pi dxi . (27)
The distribution is nonintegrable since ω ∧ dω 6= 0. In addition, ω ∧ (dω)n 6= 0 and
dim(J 1 (N ) = 2n + 1. This is exactly the local form from the Darboux theorem
and also from the local definition of the contact form. Therefore the contact space
M of dimension 2n + 1 is exactly the 1-jet of smooth functions on N .
The sections of the jet bundle s : N → J 1 (N ) are called holonomic sections
or 1-graps (in case of J k (N ) are called k-graphs) if ’differential’ coordinates are
derivatives, i.e.,
∂f
xi (s) = xi , y(s) = f, pi (s) = . (28)
∂xi
We can note that the section is the holonomic section iff it is a Legendre subman-
ifold [19]. This means that we can describe a Lagrange submanifold by a function
y(s) = f (x1 , . . . , xn ) and then all p coefficients in the Cartan distribution or P
coefficients in the contact form are derivatives

∂f
 p1 = p1 (x1 , . . . , xn ) = ∂x1

... (29)
 p = p (x , . . . , x ) = ∂f .

n n 1 n ∂xn

Symmetries of contact structure are such transformations that preserve the

Cartan distribution [12, 11]. For diffeomorphism φ : J 1 (N ) → J 1 (N ) we have the
condition that it is a contact symmetry:

φ∗ ω = λφ ω, (30)

where λφ is some smooth non-vanishing function on J 1 (N ). This condition shows

that the kernel of φ∗ ω is the same as the kernel of ω - they define the same contact
distribution.
Apart of simple symmetries like translation (xi , y, pi ) → (xi + αi , y + β, pi ) the
most important in thermodynamics is the Legendre transformation:

(xi , y, pi ) → (pi , y − xi pi , −xi ), (31)

that interchange xi with corresponding pi . The Legendre transformation can be

made only between a few variables and not the full set of them.
J 1 (N ) is sufficient for thermodynamics, however for general theory of jet spaces
consult [11, 12, 10].

17
References
[1] E. Babson, D.N. Kozlov, Group Actions on Posets, J. Algebra 285, 2, 439–450
(2005)

[2] P. Bamberg, S. Sternberg, A Course in Mathematics for Students of Physics,

Cambridge University Press, vol. 2 1990

[3] J.B. Boyling, An axiomatic approach to classical thermodynamics, Proc. R.

Soc. London, A 329, pp. 35-70 (1972)

[4] H. B. Callen, Thermodynamics, John Wiley & Sons Inc 1966

[5] T. Dieck, Transformation Groups and Representation Theory, Lecture Notes

in Mathematics 766, Springer 1979

[6] D.G.B. Edelen, Applied Exterior Calculus, Dover, 2011

[7] T. Frankel, Geometry of Physics, Cambridge UP, 2011

[8] R. Ingarden, A. Jamiolkowski, R. Mrugala, Fizyka statystyczna, PWN 1990

(in Polish)

[9] A. Katok, B. Hasselblatt, Introduction to the Modern Theory of Dynamical

Systems, Cambridge University Press; Revised edition 1996

[10] I. Kolář, P.W. Michor, J. Slovák, Natural Operations in Differential Geometry,

Springer-Verlag Berlin Heidelberg 1993

[11] A. Kushner, V. Lychagin, V. Rubtsov, Contact Geometry and Nonlinear Dif-

ferential Equations, Cambridge University Press; 1 edition 2007

[12] A. Kushner, V. Lychagin, J. Slovák, Lectures on Geometry of MongeAmpère

Equations with Maple in Nonlinear PDEs, Their Geometry, and Applications,
Birkhäuser Basel 2019

[13] R.A. Kycia, Landauers Principle as a Special Case of Galois Connection,

Entropy 2018, 20(12), 971; https://doi.org/10.3390/e20120971

[14] J. Ladyman, S. Presnell, A.J. Short, B. Groisman, The connection between

logical and thermodynamic irreversibility, Studies in History and Philosophy
of Science Part B: Studies in History and Philosophy of Modern Physics, 38, 1,
pp. 58–79 (2007); DOI: https://doi.org/10.1016/j.shpsb.2006.03.007

[15] R. Landauer, Irreversibility and heat generation in the computing process,

IBM Journal of Research and Development, 5, 183-191 (1961)

[16] E.H. Lieb, J. Yngvason, A Guide to Entropy and the Second Law of Thermo-
dynamics, Notices of The AMS, May 1998

18
[17] E. H. Lieb, J. Yngvason, The Physics and Mathematics of the Second
Law of Thermodynamics, Phys.Rept. 310 1–96 (1999); DOI: 10.1016/S0370-
1573(98)00082-9

[18] Valentin V. Lychagin, Contact Geometry, Measurement, and Thermodynam-

ics in Nonlinear PDEs, Their Geometry, and Applications, Birkhäuser Basel
2019

[19] V. V. Lychagin, Contact Geometry and nonlinear second order differential

equations, Uspechi Mat. Nauk 34, 137-165 (1979)

[20] S. Mac Lane, Categories for the Working Mathematician, Springer; 2nd ed.
(1978)

[21] O. Ore, Galois Connexions, Transactions of the American Mathematical So-

ciety 55, 493–513 (1944)

[22] F.M. Reza, An Introduction to Information Theory , Dover Publications; Re-

vised edition 1994

[23] P. Smith, Category Theory: A Gentle Introduction, Script https://www.

logicmatters.net/categories/

[24] W. Li, Y. Zhao, Q. Wang, J. Zhou, Twenty Years of Entropy Research: A

Bibliometric Overview, Entropy 21(7), 694 (2019); DOI: https://doi.org/
10.3390/e21070694

19