On solvability of linear differential equations in finite terms

Askold Khovanskii∗, and Aaron Tronsgard†

December 25, 2024

arXiv:2412.18062v1 [math.AG] 24 Dec 2024

To the memory of Andrei Bolibrukh

Abstract
We consider the problem of solvability of linear differential equations over a differential
field K. We introduce a class of special differential field extensions, which widely generalizes
the classical class of extensions of differential fields by integrals and by exponentials of integrals
and which has similar properties. We announce the following result: if a linear differential
equation over K can not be solved by generalized quadratures, then no special extension can
help solve it. In the paper we prove a weaker version of this result in which we consider
only pure transcendental extensions of K. Our paper is self-contained and understandable for
beginners. It demonstrates the power of Liouville’s original approach to problems of solvability
of equations in finite terms.

Keywords:
linear differential equations; solvability by quadratures; integration in finite terms; differential
Galois theory
2010 Math. Subj. Class. 12H20.

1 Introduction
In the 1830s Liouville started to create a theory of solvability in finite terms. In some cases it
answers the following question: is there a solution of an equation which is representable by a
certain class of formulas? In other words, is there a solution representable in finite terms?
About the same time Galois theory was invented. It provides a criterion of solvability of
algebraic equations in radicals. Liouville was inspired by Galois theory. Among his results there
are two famous theorems. The First Liouville Theorem provides a criterion of integrability of
elementary functions in elementary functions. The Second Liouville Theorem provides a criterion
of solvability of second order linear differential equations by quadratures. Modern proofs of these
theorems can be found correspondingly in [1] and [2].
Liouville’s brilliant and simple ideas were never properly understood (maybe, because Liouville
did not carefully formalize them). Only in 1910, Picard and Vessiot generalized the Second Liouville
Theorem for linear differential equations of an arbitrary order. To obtain this result they developed
a new deep differential Galois theory.
J.F. Ritt up to some extent formalized Liouville’s method and generalized it (see [3] and [4]).
Later J.F.Ritt, Kolchin and others developed differential algebra, which totally gets rid of multi-
valued analytic functions and deals with abstract differential fields. It became a well established
branch of pure algebra.
∗
The first author was partially supported by the Canadian Grant No. 156833-17.
†
The second author was partially supported by the NSERC post-graduate scholarship (PGS-D) 569475

1
 If one wants to show that a solution of an equation is not representable by any formula of a
certain type (for example is not representable by quadratures), one has to deal with the collection
of all functions representable by such formulas. Functions from this collection could be very
complicated; they could be multi-valued, they could have complicated singularities, and so on.
Such functions do not form a differential field, since arithmetic operations are not well defined for
multi-valued functions.
In the framework of differential algebra, one can state and solve the problem of solvability
of linear differential equations by quadratures. These results are applicable to analytic theory of
differential equations in the following way.
Let K be a subfield of the field M of meromorphic functions on a connected open set U on
a complex line, which is closed under differentiation. Consider a homogeneous linear differential
equation over the field K
y (n) + a1 y n−1 + · · · + an y = 0, (1)
where ai ∈ K. With a point z ∈ U which is a regular point for all coefficients ai , one can associate:
the differential field Kz of germs at z of functions from K, the differential field Fz generated over
Kz by germs at z of all solutions of (1) and the differential field Mz of all meromorphic germs at z.
Definition 1.1. An equation (1) is solvable by quadratures over K if there exists a point z ∈ U ,
such that inside the field Mz there is a chain of extensions by quadratures of the field Kz which
contains the field Fz .
One can show that the definition of solvability of an equation by quadratures which makes use
of multi-valued analytic functions is equivalent to Definition 1.1 (see [2], [5]).
Thus, results from differential algebra are applicable to the analytic theory of differential equa-
tions. Moreover, in differential algebra one can consider equations over arbitrary differential fields,
not only over a field of meromorphic functions.
On the other hand, there are operations over analytic functions which do not exist in differential
algebra. For example, one can compose functions, and it is natural to use compositions in formulas.
Problems on solvability of equations by formulas involving compositions can not be stated and
solved using a pure algebraic approach. Note, that compositions with some functions can be
defined in differential algebra (see remark 2.1).
A reduction of the Second Liouville Theorem (and its generalizations for linear differential
equations of arbitrary order) to differential algebra should, in principal, significantly simplify it: it
means that the result can be proven using only arithmetic operations and differentiation. Never-
theless, algebraic results on solvability of linear differential equations of arbitrary order, which uses
differential Galois theory (see [6]) and Rosenlicht’s algebraic proof (see [7]) of that result, which
uses valuation theory, are rather involved. In [2], the same result over functional differential fields
is proved using Liouville’s original method. This proof is relatively simple, but it uses analytic
tools and is not applicable to abstract differential fields.
Our goal is to find an algebraic proof of that result which is as simple as the original proof of
the Second Liouville’s Theorem. We are also looking for possible generalizations of that result.
We have achieved our goal. The result is announced in the next section. Currently, one step
in our proof which deals with algebraic extensions is not polished and is too involved. We have
decided in this first publication, to exclude algebraic extensions from consideration.
This makes our result weaker than the classical criterion of solvability of linear differential
equations by quadratures. But also our result in many ways is stronger than the classical criterion:
we show that many extensions (which we call special transcendental extensions) can not help to
solve linear differential equations. Besides that, our result is understandable for beginners and it
clearly demonstrates Liouville’s original approach.

2
 In subsequent publications we plan to present a proof of the announced theorem. It is not as
elementary as the present paper. We need the theory of algebraic curves over (possibly very big)
algebraically closed fields. We also plan to consider much more general special extensions which
have similar properties with special transcendental extensions.

2 Solvability over differential fields

We begin with the definition of an abstract differential field.
Definition 2.1. A differential field is a field K along with a fixed additive map K → K sending
a 7→ a′ satisfying the Leibniz rule (ab)′ = a′ b+ ab′ . The element a′ is called the derivative of a ∈ K.
Let us present several examples of differential fields.

Example 2.1.

1. The field M (U ) of all meromorphic functions on a connected open set U on the Riemann
sphere CP 1 = C1 ∪ {∞} with the natural differentiation.

2. The field of meromorphic functions M (S) on a connected Riemann surface S equipped

with non constant meromorphic projection π : S → CP 1 with the natural differentiation.

3. Any subfield of the fields M (U ) and M (S) closed under differentiation.

The differential field C(z) of rational functions of complex variable z and the differential field
C(S) of rational functions on an algebraic curve S with none constant rational projection
π : S → CP 1 belong to these examples.
An element c of a differential field K is a constant if c′ = 0. The collection of all constants in
K form the field of constants.
Throughout the paper, by a differential field we mean a differential field of characteristic 0 with
the same fixed subfield of constants. For example, one can assume that all differential fields under
consideration contain C as the subfield of constants.
Consider a nested pair K ⊂ F of differential fields.
Definition 2.2. An element y ∈ F is called

1. algebraic over K if y is a root of a polynomial over K;

2. an integral over K (or an integral of a ∈ K) if y ′ = a and a ∈ K;

3. an exponential of an integral over K (or an exponential of integral of a ∈ K) if y ′ = ay and

a ∈ K.

Remark 2.1. Many other natural definitions can be reformulated in such a way that they will
make sense over differential fields. For example, the functions u = exp f and v = ln f satisfy
the following relations: u′ = f ′ u, v ′ = f ′ /f . Thus, one can say that the element u of F is an
exponential of f ∈ K if u′ = f ′ u, and the element v of F is a logarithm of f if v ′ = f ′ /f .
Correspondingly, many different types of representability of functions in finite terms can be
reformulated for differential fields.
Let us define a notion of solvability by quadratures in differential algebra. Consider a nested
pair K ⊂ F of differential fields. Let E be an intermediate differential field, i.e., K ⊂ E ⊂ F .

3
Definition 2.3. An extension K ⊂ E is an extension by quadratures if there is a chain of field
extensions
K = K0 ⊂ · · · ⊂ Kn = E (2)
such that for each 0 ≤ i < n, the extension Ki ⊂ Ki+1 is either an extension by adjoining an
integral over Ki , or by adjoining an exponential of an integral over Ki .

1. An element a ∈ F is representable by quadratures over K if there is an extension K ⊂ E by

quadratures such that a ∈ E.

2. An equation over K is solvable by quadratures over K if there is an extension K ⊂ E by

quadratures which contains all solutions of the equation.

In the same way one defines other types of solvability of equations in differential algebra.
Definition 2.4. An extension K ⊂ E is an extension by generalized quadratures if there is a
chain (2) in which besides adjoining integrals and exponential of integrals, algebraic extensions are
allowed.

1. An element a ∈ F is representable by generalized quadratures over K if there is an extension

K ⊂ E by generalized quadratures such that a ∈ E.

2. An equation over K is solvable by generalized quadratures over K if there is an extension

K ⊂ E by generalized quadratures which contains all solutions of the equation.

Each extension Ki ⊂ Ki+1 in a chain (2) defining an extension by quadratures, or by gener-

alized quadratures is generated over Ki by one element yi . The element yi is either algebraic, or
transcendental over Ki . If yi is algebraic, one can consider the extension Ki ⊂ Ki+1 as an algebraic
equation. Otherwise, yi has to satisfy an equation yi′ = R(yi ), where R is a rational function over
Ki (see Lemma 6.3).
Definition 2.5. An element y is a special transcendental element over K if y is a transcendental
element over K and it satisfies a differential equation y ′ = R(y), where R is a rational function
over K which is divisible by y, i.e., is representable in the form R = yP
Q , where the polynomials
yP and Q are relatively prime.

Lemma 2.1. If an element y is transcendental over K and either

1. y is an exponential of integral over K,

2. y is representable as y = u−1 , where u is an integral over K,

then y is a special transcendental element over K.

Proof. 1. If y ′ = ay, then y ′ = R(y), where the rational function R(y) = ay is divisible by y;

2. if u′ = a, then y ′ = −ay 2 , i.e., y ′ = R(y), where the rational function R(y) = −ay 2 is
divisible by y.

The extension of K by u is identical to its extension by u−1 .

Definition 2.6. An extension K ⊂ E is admissible if there is a chain (2) such that each extension
Ki ⊂ Ki+1 is obtained by adjoining to Ki a special transcendental element over Ki .

4
Definition 2.7. An extension K ⊂ E is a generalized admissible extension if there is a chain (2)
such that each extension Ki ⊂ Ki+1 is either algebraic, or is obtained by adjoining to Ki a special
transcendental element over Ki .
We announce the following theorem.

Theorem 2.1 (On strong non solvability of linear differential equations). A homogeneous linear
differential equation over K can be solved by generalized admissible extensions if and only if it can
be solved by generalized quadratures.
In other words, if such an equation can not be solved by generalized quadratures, then no special
transcendental extensions can help solve it either.

In the present paper, we prove a weaker version of Theorem 2.1 which provides a criterion of
solvability of homogeneous linear differential equations over K by admissible extension. It implies
the following:

Theorem 2.2. If a homogeneous linear differential equation can be solved by admissible extensions,
then it can be solved by a nested chain of extensions by integrals and by transcendental exponentials
of integrals.

3 An illustrative example
In this section, we present a simple example which illustrates our approach and Liouville’s method.
Let K be a differential field of meromorphic functions on some connected domain on a complex
line which is closed under differentiation.
Consider the second order homogeneous equation

y ′′ + ay ′ + by = 0 (3)

over K, i.e., a = a(x) and b = b(x) are functions from K. It is well known that y is a nonzero
solution of (3) if and only if its logarithmic derivative u = y ′ /y satisfies the Riccati equation

u′ + au + b + u2 = 0. (4)

Thus, problems of solving in finite terms equations (3) and (4) are closely related, and for
solvability by quadratures are in fact equivalent. If one wants to show that the equation (3) is not
solvable by quadratures, one can try to show it instead for equation (4). Let us demonstrate how
one can make one step in proving non-solvability by quadratures of the equation (4).
Assume that there are no solutions of the Riccati equation (4) in the differential field K. Let
us show that an addition to the field K, say, an exponent y can not help in solving (4). Our goal
is to prove the following theorem.

Theorem 3.1. If an exponent y (i.e., a nonzero solution of the equation y ′ = y) is transcendental

over K, and the Riccati equation (4) has a solution in Khyi, then it has a solution in K.

Remark 3.1. Liouville’s theory of integrability in finite terms has the following slogan:
If a “simple” equation has a solution given by an “allowed” formula, then it has to have
(another) solution given by a “simple” allowed formula.
Theorem 3.1 demonstrates this slogan: here a simple equation is the equation (4), an allowed
formula represents a solution as an element of Khyi and a simple allowed formula represents a
solution as an element of K.

5
 Let us prove first two lemmas.

Lemma 3.1 (Liouville’s principle for a transcendental exponent). An element a ∈ Khyi is rep-
resentable in a unique way as a rational function in y over K. The derivative R′ of R = R(y) is
given by
R′ = Rx′ + Ry′ y, (5)
where Rx′ , Ry′ are the following differentiations of the field K(y):

1. Rx′ sends y to zero and coincides with the differentiation in K on coefficients;

2. Ry′ sends y to 1 and sends to zero all elements of the differential field K

Proof. Indeed, since y ′ = y, any element of the differential field Khyi is representable as a rational
function in y over K. If an element a can be represented as a rational function in y in two different
ways, i.e., a = R1 (y); a = R2 (y), then R1 (y) = R2 (y) which means that y is algebraic over K and
contradicts the assumption. One can check (5) using the chain rule and the identity y ′ = y.

According to Lemma 3.1, the field Khyi is isomorphic to the field K(y) of rational functions
over K with the differentiation given by (5). In the field of rational functions K(y), the order of
a rational function at the point y = 0 is defined.
Definition 3.1. A rational function R(y) over K has order m at the point y = 0 if it is repre-
sentable as R = y m R0 , where R0 is a rational function which is not divisible by any positive or
negative degree of y.
Using (5), one can check the following lemma.

Lemma 3.2. If a rational function R has order m at the point y = 0, then its derivative R′ =
Rx′ + Ry′ y has order at y = 0 not smaller than m.

Proof. If R = y m R0 , then

R′ = my m R0 + y m+1 (R0 )′y + y m (R0 )′x .

The function R0 equals to P/Q, where Q is not divisible by y. Thus, (R0 )y and (R0 )′x are equal
to (Px′ Q − P Q′x )/Q2 and (Py′ Q − P Q′y )/Q2 where polynomial Q2 is not divisible by y. So (R0 )y
and (R0 )′x have nonnegative order which implies Lemma 3.2.

Proof of Theorem 3.1. Assume that u = R(y) satisfies (4). If we plug in to the rational function
R any solution y of the equation y ′ = y, then we still will get a solution of (4) under assumption
that the plugging in of y to R is well defined.
Let us try to plug in to R the solution y ≡ 0. If we succeed, then we will have a solution R(0)
of the Riccati equation (4) in the field K. One can do it only if the order of the rational function
R(y) at the point y = 0 is nonnegative. If R has a pole at the point y = 0, then plugging y = 0 in
to R makes no sense.
Let us show that if u = R(y) has a negative order k at y = 0, then it can not satisfy equation
(4). Indeed, if u has an order k < 0, then u2 has an order 2k < 0. The terms u and u′ in (4) have
orders ≥ k. So the term u2 can not be cancelled and Riccati’s equation can not be satisfied. Thus,
if u = R satisfies (4), it can not have negative order at y = 0.
The theorem is proven.

6
4 Generalized Riccati equation
In this section, we recall classical results on the generalized Riccati equation (the presentation is
borrowed from [2]). Let y be a a nonzero element of a differential field and let u be its logarithmic
derivative, i.e., y ′ = uy.
Definition 4.1. Let Dn be a polynomial in u and in its derivatives u′ , . . . , u(n−1) up to order
(n − 1) defined inductively by the following conditions:

dDk
D0 = 1; Dk+1 = + uDk .
dx
Lemma 4.1. 1. The polynomial Dn has integral coefficients and deg Dn = n. The degree n
homogeneous part of Dn equals to un (i.e., Dn = un + D̃n , where deg D̃n < n).

2. If y is a function whose logarithmic derivative is equal to u (i.e., if y ′ = uy), then, for any
n ≥ 0, we have the relation y (n) = Dn (u)y.

Both claims of the lemma can be easily checked by induction.

Consider a homogeneous linear differential equation whose coefficients ai belong to a differential
field K:
y (n) + a1 y (n−1) + · · · + an y = 0. (6)
Definition 4.2. The equation

Dn + a1 Dn−1 + · · · + an D0 = 0 (7)

of order (n − 1) is called the generalized Riccati equation for the homogeneous linear differential
equation (6).

Lemma 4.2. A nonzero element y satisfies the linear differential equation (6) if and only if its
logarithmic derivative u = y ′ /y satisfies the corresponding generalized Riccati equation (7).

Proof. For proving Lemma 4.2 in one direction, one can divide (6) by y and use the identity
y (k) /y = Dk (u).
In the other direction, if u is a solution of (7), then multiply (7) by y and use the identity
y k = Dk (u)y. We obtain that y is a nonzero solution of (6).

5 Reduction of order
In this section, we recall the classical procedure of order reduction for homogeneous linear differ-
ential equations (the presentation is borrowed from [5]).

5.1 Division with a remainder of differential operators

A linear differential operator of order n over a differential field K is an operator L = an D n +· · ·+a0 ,
where ai ∈ K and an 6= 0, acting on an element y of any differential field F containing k by the
formula
L(y) = an y (n) + · · · + a0 y.
For operators L1 and L2 over K, their product L = L1 ◦ L2 = L1 (L2 ) is also an operator over
K. The multiplication of operators is, in general, noncommutative, but the leading term of the
operator L = L1 ◦ L2 is equal to the product of the leading terms of the operators L1 and L2 . For

7
operators L and L2 of orders n and k over K, there exist unique operators L1 and R over K such
that L = L1 ◦ L2 + R, and the order of R is strictly less than k. The operator R is called the
remainder in the right division of the operator L by the operator L2 .
The operators L1 and R can be constructed explicitly: the algorithm for the division of op-
erators with a remainder is based on the formula for the leading term of the product and is very
similar to the algorithm of the division with a remainder for polynomials in one variable.

5.2 Procedure of order reduction

y1′
Let L be a linear differential operator over K, y1 a nonzero element of the field K, p = y1 its
logarithmic derivative and L2 = D − p the first order operator annihilating y1 .

Lemma 5.1. The operator L is divisible by L2 from the right if and only if the element y1 satisfies
the identity L(y1 ) ≡ 0.

Proof. The remainder R in the right division of L by L2 is the operator of multiplication by c0 ,

where c0 = y11 L(y1 ). Indeed, the desired equality can be obtained by plugging in y = y1 to the
identity L(y) ≡ L1 ◦ L2 (y) + c0 y.

Using a nonzero solution y1 of the equation L(y) = 0 of order n, one can reduce the order of
this equation. To this end, one needs to represent the operator L in the form L = L1 ◦ L2 , where
L1 is an operator of order (n − 1).
Coefficients of the operator L1 lie in an extension of the differential field K obtained by adjoin-
ing the logarithmic derivative p of the element y1 . If one knows some solution u of the equation
L1 (u) = 0, then one can construct a certain solution y of the initial equation L(y) = 0. To do
that, it suffices to solve the equation L2 (y) = y ′ − py = u.

Lemma 5.2. An element y satisfies the equation L2 (y) = u if and only if y = y1 z, where z satisfies
the equation z ′ = u/y1 .
y1′
Proof. The operator L2 (y) = y ′ − y1 y can be rewritten in the form L2 (y) = y1 ( yy1 )′ .

The procedure just described is called the reduction of order of a differential equation.
Remark 1. An operator annihilating y1 is defined up to a factor, which can be an arbitrary
element of the field K, and the procedure of the order reduction depends on the choice of this
element. It is easier to divide L from the right by the operator L̃2 = D ◦ y1−1 and represent L
in the form L = L̃1 L̃2 which reduces the initial equation L(y) = 0 to the equation L̃1 (u) = 0 of
smaller order. Usually, this very procedure of order reduction is given in textbooks on differential
equations. Note that the coefficients of the operator L̃1 lie in the extension of the differential field
K obtained by adjoining the element y1 itself, rather than its logarithmic derivative p. This makes
the operator L̃1 inconvenient for the purposes of our paper.

6 Pure transcendental extensions

A field extension K ⊂ F is pure transcendental if any element y ∈ F , which is algebraic over K,
belongs to K. Let us start with the following example. Let y be an integral over K, i.e., y ′ = a
for some a ∈ K.

Lemma 6.1. The integral y either belongs to the field K or is transcendental over K.

8
Proof. By an assumption, all differential fields under consideration contains the same field of
constants. Thus, if y ′ = 0, then y belongs to K. Assume that a 6= 0, and its integral y is algebraic
over the field K. Let Q be a nonzero polynomial over K of the smallest possible degree such that
Q(y) = an y n · · · + a0 = 0.
We can assume that n > 1 and that an = 1. Differentiating the identity Q(y) = 0, we obtain
the equation nay n−1 + · · · + a′0 = 0 on y of a smaller degree. This contradicts our choice of the
polynomial Q.

Lemma 6.1 implies that any nontrivial extension by adjoining an integral is pure transcendental.
Let us describe all pure transcendental extensions of transcendental degree one.
Let K ⊂ F be a nested pair of differential fields. Assume that y ∈ F is transcendental over
K and the algebraic field K(y) ⊂ F generated by y and K is closed under differentiation, in
particular, y ′ belongs to the algebraic field K(y), i.e., y ′ = R(y), where R is some rational function
over K.
Liouville’s principle shows that for any R ∈ K(y) there is a pure transcendental extension
K ⊂ Khyi in which the identity y ′ = R(y) holds and the rational function R together with
differentiation in the field K completely determine the differential field Khyi.
Lemma 6.2. Let K be a differential field, and K(t) ⊃ K the field of rational functions over K
in indeterminate t. Then for any choice of rational function R ∈ K(t), there exists a unique
differentiation on the field K(t) ⊃ K which coincides with the differentiation on the subfield K
and is equal to R(t) on t.
Proof. Consider the map sending R ∈ K(t) to R′ ∈ K(t) given by the formula

R ′ = RK
′
+ RRt , (8)

where RK′ denotes the derivative of R by the differentiation rule of K, viewing t as a constant. And

Rt denote the usual partial derivative with respect to t, viewing the elements of K as constant.
One can check the the map R → R′ provides a differentiation of the field K(t) with satisfies all
needed conditions. Such differentiation is unique, since K(t) is generated by t and K.

Lemma 6.3 (Liouville’s Principle). Let F be some differential field extension of K, and y ∈ F
transcendental over K. Suppose that y satisfies an equation y ′ = R(y), where R is a rational
function with coefficients in K. Then Khyi is isomorphic to the field K(t) of rational functions
over K with differentiation given t′ = R(t) (see Lemma 6.2)
Proof. Consider the map from the field K(t) of rational functions over K to the field Khyi which
fixes elements of K and sends t to y. Since y is transcendental over K, this map is injective.
Indeed, if R1 (y) = R2 (y) for different functions R1 , R2 ∈ K(t) then the identity R1 (y) = R2 (y)
provides an algebraic equation over K on y which is impossible.
So, the field Khyi contains an isomorphic copy of (the algebraic field) K(t), and by Lemma 6.2
K(t) can be made a differential field with the choice t′ = R(t). Our map preserves differentiation
between the differential field K(t) and its image in Khyi. And since Khyi is the smallest differential
field containing K and y, we must have that Khyi is precisely the image of K(t).

The following lemma is obvious.

Lemma 6.4. Let K = K0 ⊂ · · · ⊂ Kn = F be a chain of fields. If for any i = 0, . . . , n − 1 the
extension Ki ⊂ Ki+1 is pure transcendental, then the extension K ⊂ F also is pure transcendental.
Proof. Assume that y ∈ Ki+1 is algebraic over K0 . Then since K0 ⊂ Ki , the element y belongs to
Ki . By repeating this argument, one obtains that y ∈ K0 .

9
 Consider a homogeneous linear differential equation (6) over the field C(z) of complex rational
functions with the usual differentiation.

Lemma 6.5. If all coefficients of the equation (6) are polynomials ai ∈ C[z], then the extension
M of the field C(z), obtained by adjoining all solutions of (6), is pure transcendental.

Proof. The extension M is contained in the field of meromorphic functions on the complex line.
A meromorphic function is algebraic over C(z) if and only if it is rational.

7 Adjoining a special transcendental element

In this section, we prove Lemma 7.2 which provides an important property of special transcendental
extensions.
For a rational function R ∈ K(y) over a field K, we denote by ord R its order at the point
y = 0 (see Definition 3.1). For any functions R1 , R2 ∈ K(y), the following relations hold:

1. ord (R1 R2 ) = ord R1 + ord R2

2. ord (R1 + R2 ) ≥ min(ord R1 , ord R2 ), and if ord R1 ) 6= ord R2 ,

then ord (R1 + R2 ) = min(ord R1 , ord R2 ).

Let D : K(y) → K(y) be any differentiation of the field K(y).

Lemma 7.1. Nonzero elements f ∈ K(y) satisfying inequality ord D(f ) ≥ ord f form a multi-
plicative subgroup in the field K(y).

Proof. The inequality ord D(f ) ≥ ord f can be rewritten as ord (Df /f ) ≥ 0. The logarithmic
derivative Df /f provides a homomorphism of the multiplicative group of the field K(y) to its
additive group. Elements f satisfy ord Df ≥ ord f if and only if it belongs to the pre-image of the
additive subgroup of functions g ∈ K(y) having nonnegative order ord g ≥ 0.

Let y be a special transcendental element over a differential field K satisfying the differential
equation y ′ = R(y) = yR0 (y), where R0 has nonnegative order at y = 0, i.e., ord R0 ≥ 0. The
differential field Khyi is isomorphic to the field K(y) with the differentiation R′ = RK
′ + yR R .
0 y

Lemma 7.2. For any element R in the differential field Khyi, the following inequality holds:

ord R′ ≥ ord R (9)

Proof. Indeed, R′ = RK ′ + yR R′ . It is easy to check that ord R′ ≥ ord R and ord yR′ ≥ ord R.
0 y K y
Indeed, these inequalities obviously hold for polynomials over K. Now one can apply Lemma 7.1,
and the lemma follows from the properties of order listed above.

8 Existence of a solution in an admissible extension

In this section, we prove the following theorem.

Theorem 8.1. A homogeneous linear differential equation (6) has a solution in some admissible
extension of the field K if and only if it has a solution y whose logarithmic derivative belongs to
K, and y either belongs to K, or is transcendental over K.

The proof is based on the following lemma.

10
Lemma 8.1. Let K ⊂ Khyi be an extension by adjoining a special transcendental element y over
K. The generalized Riccati equation (7) for the equation (6) has a solution in Khyi if and only if
it has a solution in the field K.

Let T be a polynomial over K in u and in its derivatives u, u′ , . . . , u(k) , . . . .

Definition 8.1. Say that T of some degree n is a Rosenlicht type polynomial if the degree n
homogeneous part of T is equal to un (i.e., T = un + T̃ , where deg T̃ < n).
An equation
T (u, u′ , . . . , u(n) ) = 0, (10)
where T is a Rosenlicht type polynomial, is called a Rosenlicht type equation.
The generalized Riccati equation is a Rosenlicht type equation. This property of the generalized
Riccati equation plays the key role in the classical approach (which goes back to Liouville and was
specified by Rosenlicht) to the problem of solvability of homogeneous linear differential equations
in finite terms. Thus, instead of Lemma 8.1, we prove the following more general lemma.

Lemma 8.2. Let K ⊂ Khyi be an extension by adjoining a special transcendental element y over
K. An equation (10) has a solution in Khyi if and only if it has a solution in the field K.

Proof. Let R(y) ∈ Khyi where y is a special transcendental element over K. Let us show that if
R(y) satisfies (10), then ord R = m is nonnegative. Indeed, if m < 0 then ord un = nm is strictly
smaller than ord T̃ , which follows from the definition of Rosenlicht polynomial and Lemma 7.2.
Thus, m ≥ 0. So, one can evaluate both sides of (10) at y=0 and obtain a solution of (10) in the
field K.

Corollary 8.1. An equation (10) has a solution in an admissible extension E of the field K if
and only if it has a solution in the field K.

Proof. Let K0 ⊂ . . . , ⊂ Kn = E be a chain in which each extension Ki ⊂ Ki+1 is obtained by

adjoining a special transcendental element over Ki . By Lemma 8.2, if (10) has a solution in Ki+1 ,
it also has a solution in Ki . This argument proves corollary.

Proof of Theorem 8.1. By Lemma 4.2, the logarithmic derivative of any nonzero solution y of (6)
has to satisfy its generalized Riccati equation. Assume that y belongs to an admissible extension
E of K. Corollary 8.1 implies that the generalized Riccati equation has a solution u ∈ K. Thus,
(6) has a nonzero solution z such that z ′ /z = u ∈ K. If this element z is algebraic over K, it
belongs to K, since E is a pure transcendental extension of K.

9 Algebraic relations between exponentials of integrals

Let K be a differential field. For y belonging to the multiplicative group K ∗ = K \ {0} of the field
K, the logarithmic derivative y ′ /y is defined. It is easy to check the following lemma.

Lemma 9.1. The map, which sends each element y ∈ K ∗ to its logarithmic derivative, provides
a homomorphism of the multiplicative group K ∗ to the additive group K+ of the field K, i.e., for
any u, v ∈ K ∗ any integers k, n the following identity holds:

(uk v n )′ u′ v′
= k + n .
uk v n u v

11
Definition 9.1. Let F be any extension of a differential field K. A nonzero element y ∈ F is an
exponential of integral of a ∈ K if y ′ = ay. In other words, y ∈ F ∗ is an exponential of integral of
a ∈ k if a is the logarithmic derivative of y.
Let F be an extension of K.

Lemma 9.2. Assume that a nonzero element y ∈ F satisfies an equation y ′ = ay for a ∈ K and
an element u ∈ F is representable in a form u = cy m , where c ∈ K, m ∈ Z. Then

1. the derivative u′ of u also is representable in the form u′ = c̃y m , where c̃ = c′ + ma;

2. u′ = 0 if and only if c = λy −m , where λ is a constant, i.e., λ′ = 0. If u′ = 0, then y m ∈ K.

Proof. The first relation follows from direct differentiation. If u′ = 0, then u is a constant λ. Thus,
cy m = λ which implies the relation c = λy −m and y m = c−1 λ.

Definition 9.2. Let F = Khyi be an extension of a differential field K by an element y. Then F

is an extension by exponential of integral of a ∈ K (a transcendental extension by exponential of
integral of a ∈ K) if y ′ = ay (if y is transcendental over K and y ′ = ay).

Theorem 9.1. Let y1 , . . . , yn ∈ F be the exponentials of integrals of elements a1 , . . . , an ∈ K,

where K is a differential subfield of F . Assume that there are no nontrivial monomials u =
y1k1 . . . ynkn , where k1 , . . . , kn are not necessary positive integers (assuming that, at least, one ki 6= 0)
such that a relation u = c for some c ∈ K holds. Then the elements y1 , . . . , yn are algebraically
independent over the field K.
P
Proof. Assume that y1 , . . . , yn are algebraically dependent and let P = cm1 ,...,mn xm mn
1 . . . xn be
1

a nonzero Laurent polynomial over K (i.e., P m1 , . . . , mmn are not necessarily positive integral num-
bers) which annihilates y1 , . . . , yn , i.e., cm1 ,...,mn y1 1 . . . ynmn = 0. One can choose such Laurent
polynomial P which contains the smallest possible number of nonzero terms cm1 ,...,mn y1m1 . . . ynmn .
By dividing the relation P = 0 by one of its terms, one obtains a Laurent polynomial P̃ having
the same number of terms as P , but one of its terms is equal to 1. By Lemma 9.2, the derivative
u′ of term u = cm1 ,...,mn y1m1 . . . ynmn has a form u′ = c̃m1 ,...,mn y1m1 . . . ynmn for some c̃m1 ,...,mn ∈ K.
Thus, by differentiating the identity P̃ = 0, one obtains a relation P̃ ′ (y1 , . . . , yn ) = 0 containing
smaller number of terms (the derivative of the term 1 is equal to zero). Thus, the derivative of each
term of the Laurent polynomial P̃ is equal to zero. By Lemma 9.2, it means that some nontrivial
monomial y1m1 . . . ynmn belongs to the field K.

10 Transcendental extensions by exponentials of integrals

Assume that a differential field F = Khy1 , . . . , yn i is generated over a field K by nonzero elements
y1 , . . . , yn , such that yi′ = ai yi , where ai belong to K. Let G be the multiplicative subgroup in
F ∗ generated by elements y1 , . . . , yn , and let G0 be its subgroup which is equal to the intersection
of G with the field K, G0 = G ∩ K ∗ . Denote by Ψ0 ⊂ Ψ the images of the multiplicative groups
G0 , G under the map which sends y ∈ F ∗ to y ′ /y ∈ F+ . By construction, the group Ψ is generated
by the elements ai ∈ K+ .

Theorem 10.1. The factor group G/G0 has the following properties:

1. If G/G0 is torsion free, then there is a chain of fields K = K0 ⊂ · · · ⊂ Km = F , such that

for i = 0, . . . , m − 1 the field Ki+1 is equal to Ki hui+1 i, where ui+1 is an exponent of an
integral over Ki , transcendental over Ki .

12
 2. If G/G0 has nontrivial torsion, then there is no chain of fields K = K0 ⊂ · · · ⊂ Km ⊃ F ,
such that for i = 0, . . . , m − 1 the field Ki+1 is a pure transcendental extension of the field
Ki .

Proof. 1. Let u1 , . . . , um be generators of the factor group G/G0 . The elements u1 , . . . , um can
be chosen to be exponentials of integrals over the field K0 . Each monomial in u1 , . . . , um
does not belong to the subgroup G0 = G ∩ K0 , since the factor group is torsion free. Thus,
by Theorem 9.1, elements u1 , . . . , um are algebraically independent over K0 . Let Ki for i =
1, . . . , m be the field K0 hu1 , . . . , ui i. By construction in the chain of fields, K0 ⊂ K1 · · · ⊂ Km .
for i = 0, . . . , m − 1 the field Ki+1 is the extension of Ki by exponent of integral ui+1 over
the field K0 ⊂ Ki . The field Km is equal to the field F . All these extensions Ki ⊂ Ki + 1
are pure transcendental, since u1 , . . . , um are algebraically independent over K0 .

2. If the group G/G0 has nontrivial torsion, then there is a monomial u in y1 , . . . , yn which does
not belong no G0 but some positive power of it does. Thus, u does not belong to the field
K0 but some positive power of it does.
The monomial u is an algebraic element over K0 which does not belong to K0 . Any chain
K0 ⊂ · · · ⊂ Km of pure transcendental extensions provides a transcendental extension K0 ⊂
Km which can not contain the element u ∈ F .

11 Criterion of solvability in admissible extensions

Let L = an D n + . . . + a0 , where ai ∈ K be a linear differential operator of order n over K. In this
section, we will state and prove a criterion of solvability of the homogeneous equation

L(y) = an y (n) + · · · + a1 y = 0 (11)

over K by an admissible extension E of K.

Theorem 11.1. All solutions of (11) belong to an admissible extension E ⊃ K if and only if

1. there is a sequence of nonzero elements z1 , . . . , zn which are exponential of integrals over K,

i.e., z1′ = p1 z1 , . . . , zn′ = pn zn and p1 , . . . , pn ∈ K such that the operator L is representable
in the form
L = an L n ◦ · · · ◦ L 1 , (12)
where each operator Li is an operator of order one. Moreover, Li is equal to D − pi ;

2. let G be the multiplicative group generated by z1 , . . . , zn and let G0 be the intersection of G

with K, G0 = G ∩ K. Then the factor group G/G0 is torsion free.

Proof. If there is an admissible extension E containing all solutions of (12), then by Theorem 8.1,
the equation has to have a solution z1 whose logarithmic derivative p1 belongs to K. Let L1 be
the operator y ′ − p1 y1 . Let us divide L on L1 from the right, L = L̃1 ◦ L1 . The coefficients of L̃1
belongs to K, since p1 ∈ K. Each solution u of L̃1 (y) = 0 is representable in the form u = L1 (y),
where y is a solution of the original equation. So all solutions of L̃1 (y) = 0 also belong to the
admissible extension E of the field K.
Thus, one can apply Theorem 8.1 to the equation L̃1 (y) = 0 to find its solution z2 whose
logarithmic derivative p2 belongs to K. Let L2 be the operator D − p2 and let us represent L̃1 in
the form L̃1 = L̃2 ◦ L2 .

13
 Repeating this process, one obtains a sequence of exponentials of integrals of p1 , . . . , pn ∈ K
for which decomposition (12) of the operator L holds.
The exponential of integrals z1 , . . . , zn belong to an admissible extension E. By Theorem 10.1,
the group G/G0 is torsion free.
The criterion for solvability is proven in one direction. Let us prove it in the opposite direction.
If the group G/G0 is torsion free, by Theorem 10.1, one can construct an admissible extension E1
of K which contains all the exponential of integrals z1 , . . . , zn . Over the field E1 the operator L
is decomposed as a product of factors D − pi . Thus, the equation L = 0 one can solved step by
step, solving equations Li (y) = u. Since zi ∈ E, solutions of these equation can be obtained in
extensions by integrals (see Lemma 5.2). By Lemma 6.1, such extensions are pure transcendental.
So, extending E1 by a chain of extensions obtained by adjoining integrals one at a time, one obtains
an admissible extension E which contains all solutions of (11).

Note that if the group G/G0 has nontrivial torsion, then the extension of K by adjoining all
solutions of (11) is not pure transcendental. If it is pure transcendental (as in Lemma 6.5), then
the group G/G0 automatically has no torsion. Such an equation is solvable by admissible extension
if and only if the operator L admits a decomposition (12).

References
[1] Khovanskii, A. Integrability in finite terms and actions of Lie groups. Moscow Math-
ematical Journal, 19(2):329–341. 2019.

[2] Khovanskii, A. Solvability of equations by quadratures and Newton’s theorem. Arnold

Mathematical Journal, 4(2):193-211. 2018.

[3] Ritt, J.F. Integration in Finite Terms. Liouville’s Theory of Elementary Methods.
Columbia University Press, New York, NY. 1948.

[4] Khovanskii, A. Comments on JF Ritt’s Book “Integration in Finite Terms”. in

book: Integration in Finite Terms: Fundamental Sources by Clemens Raab (Edi-
tor), Michael F. Singer (Editor). Springer. Series: Text & Monographs in Symbolic
Computation. 2022. 137-201.

[5] Khovanskii, A. Topological Galois Theory. Springer-Verlag, Heidelberg. 2014.

[6] Marius van der Put and Michael Singer. Galois theory of linear differential equations,
volume 328 of Grundlehren der Mathematischen Wissenschaften [Fundamental Prin-
ciples of Mathematical Sciences]. Springer-Verlag, Berlin. 2003.

[7] Rosenlicht, M. An analogue of l’Hospital’s rule. Proc. Amer. Math. Soc. 37:369-373.
1973.

Askold Khovanskii
Department of Mathematics, University of Toronto, Toronto, ON, Canada
email: [email protected]

Aaron Tronsgard
Department of Mathematics, University of Toronto, Toronto, ON, Canada
email: [email protected]

14