Simple Symbolic Mathematics for Scala
This project contains a very simple, and incomplete, symbolic math library in Scala. It can differentiate and evaluate simple mathematical expressions. The library also contains some aspects of an internal DSL: The expressions can be entered like regular math with Int or Double numbers, and there is a ML style "let" expression. Here is a short example that demonstrates the differentiation feature:
import symathm.Expression._ import symathm.ExprOps._ //Create some symbols (unknown variables). val (a, x) = (Sym("a"), Sym("x")) //Create an expression. `~^` denotes exponentiation (power). val expr1 = a * x~^4 + 5 * x~^2 + x~^0.5 //Differentiate the expression with respect to `x`. val dexpr1 = diff(expr1, x) //Print the expression in human readable form. //Prints: "4.0 * a * x ~^ 3.0 + 10.0 * x + 0.5 * x ~^ -0.5;;" pprintln(dexpr1)
The library is not intended to be used seriously. Instead it should demonstrate simple features of Scala that are interesting for programmers that come form traditional object oriented languages; such as: C++, Java, Python, Ruby. The project should especially demonstrate the usefulness of pattern matching. Therefore this library is implemented three times with different programming paradigms, but with identical features and interfaces:
|Package symathm :||Functional, with pattern matching.|
|Package symathv :||Object oriented with Visitor pattern.|
|Package symathoo:||Classical object oriented.|
The three libraries are big enough (500 to 700 lines) to give an impression how working with a real program would be. But they are small and simple enough, to be easily understood. To write the algorithms, and to verify their correctness, only high school math is necessary. In principle the algorithms can be looked up in Wikipedia (http://en.wikipedia.org/wiki/Table_of_derivatives).
- Symbolic math library, implemented in functional fashion, with pattern matching. (Package: symathm)
- Implementation of the library with the visitor pattern. Object oriented, but structure similar to pattern matching. (Package: symathv)
- The symbolic math library implemented in simple object oriented fashion. (Package: symathoo)
Program that demonstrates the features of the libraries.
The three implementations of the library have identical features and interfaces.
- Short example implementation of the libraries' "DSL" features.
- Short example implementation of the visitor pattern.
- Compile all Scala source files.
- Create API documentation with scaladoc.
- Run all tests. Compiles all sources prior to testing.
- Project file for the Eclipse IDE and its Scala plugin. Lets Eclipse find the Scala files.
- This file.
Getting the Software
Either get the software by cloning the repository with Mercurial:
hg clone https://bitbucket.org/eike_welk/scala-symbolic-algebra-test
Or download (and extract) one of the auto-generated archives from here:
Run the script make-compile.sh to compile all source files. This might take a minute or two:
Then run any object with a main method. Start with the usage example UseTheLibraries, which explains all of the libraries' (few) features:
scala -classpath bin/ UseTheLibraries
To start Scala's read-eval-print loop, you need to specify the classpath where the compiled files are found (but don't specify an object that should be run):
scala -classpath bin/
With Eclipse IDE
The root directory contains a project file (.project) for the Sala IDE for Eclipse. It lets Eclipse find the source files, so that it can compile them automatically. You can run any file that contains an object with a main method by clicking the Run button.
First try out the usage example UseTheLibraries.scala.
Sala IDE for Eclipse can be downloaded from: http://scala-ide.org/
Required Knowledge of Scala
Only little knowledge of Scala is needed to understand the code. A good introductory text on Scala is:
The text above unfortunately does not cover pattern matching, which is IMHO one of Scala's main attractions. Pattern matching is covered here:
Either a working Scala installation (programs scalac and scala) on a Unix-like operating system. (On Windows you have to come up with the right command to compile the sources and run them yourself.)
Or even better a IDE with Scala support, for example the Scala-IDE for Eclipse.
To compare the characteristics of the different programming paradigms, you can add features to each version of the library.
- Add derivation of the Log node.
- Add new nodes, for example sin, cos and tan.
- Add function call node. Maybe this makes an inert diff node superfluous. (See point below.)
- Add lambda (function body) node.
- Implement an inert diff node. The "a$x" notation is a hack.
- Implement a node for a for loop. Write evaluation and differentiation algorithms for it. (I believe differentiating a for loop is possible, because older versions of Maple could do it.)
- Implement an algorithm to distribute factors over sums, and distribute powers over products. For example: (a + b) * c --> a*c + b*c. This is interesting for eval: more operators with only numeric arguments can be found, and evaluated.
- Implement an algorithm to collect factors and powers. (The opposite of the algorithm above.) It makes formulas look good.
- Maybe add a separate simplify function.
- Implement some of the TODOs in the code.
All important data structures are defined in object Expression.
Mathematical formulas are internally represented as nested trees of nodes. They are implemented as case classes, syntactical sugar for simple classes that are intended to work with the match statement. (http://www.artima.com/pins1ed/case-classes-and-pattern-matching.html)
- Expr : The common base class of all nodes
- Num(num: Double) : A number (floating point)
- Sym(name: String) : A variable (symbol)
- Add(summands: List[Expr]) : Addition (n-ary)
- Mul(factors: List[Expr]) : Multiplication (n-ary)
- Pow(base: Expr, exponent: Expr): Exponentiation (operator ~^)
- Log(base: Expr, power: Expr) : Logarithm
- Let(name: String, value: Expr, exprNext: Expr): Bind a value to a variable, and put a single expression into the environment, where the new variables are visible.
- Asg(lhs: Expr, rhs: Expr) : := operator, helper object to create Let nodes.
1+a and 1+a*2 are respectively expressed as:
Add(List(Num(1.0), Sym("a"))) Add(List(Num(1.0), Mul(List(Sym("a"), Num(2.0)))))
N-Ary Addition and Multiplication
Addition and multiplication are n-ary, they can have an arbitrary number of arguments. 1 + a + 2 + 3 and 1 * a * 2 * 3 are respectively expressed as:
Add(List(Num(1.0), Sym("a"), Num(2.0), Num(3.0))) Mul(List(Num(1.0), Sym("a"), Num(2.0), Num(3.0)))
There are no nodes for subtraction or division. Subtraction is represented as multiplication with -1: (-a = -1 * a). Division is expressed as a power of -1: (1/a = a~^(-1)). .
As there are no subtraction or division operators, a-x and a/x are respectively expressed as:
Add(List(Sym("a"), Mul(List(Num(-1.0), Sym("x"))))) Mul(List(Sym("a"), Pow(Sym("x"), Num(-1.0))))
Let nodes are similar to assignment statements in imperative programming languages. They are commands to create a new environment, where a variable is bound to a value. The dependent expression (the third argument of Let) is evaluated in this new environment. The value of a Let node is the value of its dependent expression.
Let nodes are created by a little abuse of Scala's liberal syntax (the DSL): let (a:=2) in a*a results in:
Let("a", Num(2.0), Mul(List(Sym("a"), Sym("a"))))
The Let node does not create the new environment by itself, it is interpreted by an algorithm. The eval algorithm interprets Let nodes as described above: It creates a new environment that contains the new variables and also the variables of the old environment. Then eval evaluates the dependent expression in the new environment. The expression above would be evaluated to Num(4).
The environment that stores the bindings between variables and their values, is (currently) implemented as a Map[String, Expr].
For convenience there are type (and companion object) Environment, and a call-able object Env to create environments.
The library contains a modest attempt to implement a domain specific language (DSL). The implementation of the DSL is in object Expression. Additionally there is a small example program to illustrate the same technique: src/pattern/testdsl.scala.
The common base class of all nodes, Expr, contains the usual mathematical operators: + - * / ~^, and additionally :=. (~^ is the exponentiation operator.) Each operator returns a part of the tree. The + operator, for example, returns an Add node.
There are implicit conversion functions (int2Num, double2Num) in Expression, that convert Int and Double objects into Num nodes. This way numbers and nodes can be freely mixed.
Let nodes can be somewhat elegantly created with the call-able helper object let. When called with multiple assignments, let creates nested Let nodes. The syntax is:
let (a := 2, x := 3) in a + x
Let("a", Num(2.0), Let("x", Num(3.0), Add(List(Sym("a"), Sym("x")))))
The eval algorithm would evaluate this expression to Num(5).
The high-level algorithms are implemented in object ExprOps (in all implementations of the library). All algorithms traverse a tree of nodes in a recursive way, and create a new tree as the result.
There are currently two algorithms:
This algorithm behaves like an interpreter of a programming language.
Differently to a traditional programming language there are no "unknown variable" errors. The algorithm replaces known variables (symbols) by their values, but unknown variables are left unchanged.
eval performs the usual arithmetic operations on numbers. Expressions that contain numbers and unknown variables are simplified as much as possible. The n-ary multiplication and addition nodes simplify this task: 1 + a + 2 can easily be simplified to 3 + a, by partitioning the summands into numbers and other nodes.
The algorithm can differentiate Let nodes; differentiating:
let (a:=f) in g
with respect to x, basically yields:
let (a:=f, a$x:=diff(f, x)) in diff(g, x)
|||The ideas for n-ary operators (additon, multiplication), and for the ommission of subtraction and division nodes, were taken from the computer algebra program Maxima. It is intended to simplify the algorithms.|