1

Taming math and physics using SymPy

Tutorial based on the No bullshit guide series of textbooks by Ivan Savov

Abstract—Most people consider math and physics to be scary Using SymPy

beasts from which it is best to keep one’s distance. Computers,
however, can help us tame the complexity and tedious arithmetic The easiest way to use SymPy, provided you’re connected to the
manipulations associated with these subjects. Indeed, math and internet, is to visit http://live.sympy.org. You’ll be presented with
physics are much more approachable once you have the power of
computers on your side. an interactive prompt into which you can enter your commands—right
in your browser.
This tutorial serves a dual purpose. On one hand, it serves
as a review of the fundamental concepts of mathematics for If you want to use SymPy on your own computer, you must install
computer-literate people. On the other hand, this tutorial serves Python and the python package sympy. You can then open a command
to demonstrate to students how a computer algebra system can
help them with their classwork. A word of warning is in order. prompt and start a SymPy session using:
Please don’t use SymPy to avoid the suffering associated with your you@host$ python
homework! Teachers assign homework problems to you because Python X.Y.Z
they want you to learn. Do your homework by hand, but if you [GCC a.b.c (Build Info)] on platform
want, you can check your answers using SymPy. Better yet, use Type "help", "copyright", or "license" for more information.
SymPy to invent extra practice problems for yourself. >>> from sympy import *
>>>
Contents
The >>> prompt indicates you’re in the Python shell which accepts
Python commands. The command from sympy import * imports all
I Fundamentals of mathematics 1
the SymPy functions into the current namespace. All SymPy functions
II Complex numbers 4 are now available to you. To exit the python shell press CTRL+D.
I highly recommend you also install ipython, which is an improved
III Calculus 5 interactive python shell. If you have ipython and SymPy installed,
you can start an ipython shell with SymPy pre-imported using the
IV Vectors 8 command isympy. For an even better experience, you can try jupyter
notebook, which is a web frontend for the ipython shell.
V Mechanics 9
Each section of this tutorial begins with a python import statement
VI Linear algebra 10 for the functions used in that section. If you use the statement from
Introduction sympy import * in the beginning of your code, you don’t need to
run these individual import statements, but I’ve included them so
You can use a computer algebra system (CAS) to compute compli- you’ll know which SymPy vocabulary is covered in each section.
cated math expressions, solve equations, perform calculus procedures,
and simulate physics systems.
I. Fundamentals of mathematics
All computer algebra systems offer essentially the same function-
ality, so it doesn’t matter which system you use: there are free Let’s begin by learning about the basic SymPy objects and the
systems like SymPy, Magma, or Octave, and commercial systems like operations we can carry out on them. We’ll learn the SymPy equivalents
Maple, MATLAB, and Mathematica. This tutorial is an introduction to of many math verbs like “to solve” (an equation), “to expand” (an
SymPy, which is a symbolic computer algebra system written in the expression), “to factor” (a polynomial).
programming language Python. In a symbolic CAS, numbers and
operations are represented symbolically,
√ so the answers obtained are
Numbers
exact. For example, the number 2 is represented in SymPy as the
object Pow(2,1/2), whereas >>> from sympy import sympify, S, evalf, N
√ in numerical computer algebra systems
like Octave, the number 2 is represented as the approximation In Python, there are two types of number objects: ints and floats.
1.41421356237310 (a float). For most purposes the approxima-
>>> 3
tion is okay, but sometimes approximations can lead to problems:
3 # an int
float(sqrt(2))*float(sqrt(2)) = 2.00000000000000044 = 6 2. Be- >>> 3.0
cause SymPy uses exact representations, you’ll never run into such 3.0 # a float
problems: Pow(2,1/2)*Pow(2,1/2)= 2.
Integer objects in Python are a faithful representation of the set of
This tutorial is organized as follows. We’ll begin by introducing the integers Z = {. . . , −2, −1, 0, 1, 2, . . .}. Floating point numbers are
SymPy basics and the bread-and-butter functions used for manipulat- approximate representations of the reals R. Regardless of its absolute
ing expressions and solving equations. Afterward, we’ll discuss the size, a floating point number is only accurate to 16 decimals.
SymPy functions that implement calculus operations like differentiation
and integration. We’ll also introduce the functions used to deal with Special care is required when specifying rational numbers, because
vectors and complex numbers. Later we’ll see how to use vectors and integer division might not produce the answer you want. In other
words, Python will not automatically convert the answer to a floating
integrals to understand Newtonian mechanics. In the last section,
point number, but instead round the answer to the closest integer:
we’ll introduce the linear algebra functions available in SymPy.
>>> 1/7
This tutorial presents many explanations as blocks of code. Be sure 0 # int/int gives int
to try the code examples on your own by typing the commands into To avoid this problem, you can force float division by using the
SymPy. It’s always important to verify for yourself! number 1.0 instead of 1:
 2

>>> 1.0/7 p + 2 # = Add(Symbol('p'), Integer(2))

0.14285714285714285 # float/int gives float
You can define a sequence of variables using the following notation:
This result is better, but it’s still only an approximation of the exact >>> a0, a1, a2, a3 = symbols('a0:4')
number 17 ∈ Q, since a float has 16 decimals while the decimal
expansion of 17 is infinitely long. To obtain an exact representation You can use any name you want for a variable, but it’s best if you
of 17 you need to create a SymPy expression. You can sympify any avoid the letters Q,C,O,S,I,N and E because they
√ have special uses
expression using the shortcut function S(): in SymPy: I is the unit imaginary number i ≡ −1, E is the base of
S('1/7') the natural logarithm, S() is the sympify function, N() is used to
1/7 # = Rational(1,7) obtain numeric approximations, and O is used for big-O notation.
Note the input to S() is specified as a text string delimited by quotes. The underscore symbol _ is a special variable that contains the result
We could have achieved the same result using S(’1’)/7 since a SymPy of the last printed value. The variable _ is analogous to the ans button
object divided by an int is a SymPy object. on certain calculators, and is useful in multi-step calculations:
>>> 3+3
Except for the tricky Python division operator, other math operators
6
like addition +, subtraction -, and multiplication * work as you would
>>> _*2
expect. The syntax ** is used in Python to denote exponentiation:
12
>>> 2**10 # same as S('2^10')
1024

When solving math problems, it’s best to work with SymPy objects, Expressions
and wait to compute the numeric answer in the end. To obtain a >>> from sympy import simplify, factor, expand, collect
numeric approximation of a SymPy object as a float, call its .evalf()
You define SymPy expressions by combining symbols with basic math
method:
operations and other functions:
>>> pi
>>> expr = 2*x + 3*x - sin(x) - 3*x + 42
pi
>>> simplify(expr)
>>> pi.evalf()
2*x - sin(x) + 42
3.14159265358979
The function simplify can be used on any expression to simplify
The method .n() is equivalent to .evalf(). The global SymPy
it. The examples below illustrate other useful SymPy functions that
function N() can also be used to to compute numerical values. You can
correspond to common mathematical operations on expressions:
easily change the number of digits of precision of the approximation.
Enter pi.n(400) to obtain an approximation of π to 400 decimals. >>> factor( x**2-2*x-8 )
(x - 4)*(x + 2)
>>> expand( (x-4)*(x+2) )
x**2 - 2*x - 8
Symbols
>>> collect(x**2 + x*b + a*x + a*b, x)
>>> from sympy import Symbol, symbols x**2 + (a+b)*x + a*b # collect terms for diff. pows of x
Python is a civilized language so there’s no need to define variables To substitute a given value into an expression, call the .subs()
before assigning values to them. When you write a = 3, you define a method, passing in a python dictionary object { key:val, ... }
new name a and set it to the value 3. You can now use the name a with the symbol–value substitutions you want to make:
in subsequent calculations. >>> expr = sin(x) + cos(y)
>>> expr
Most interesting SymPy calculations require us to define symbols, sin(x) + cos(y)
which are the SymPy objects for representing variables and unknowns. >>> expr.subs({x:1, y:2})
For your convenience, when live.sympy.org starts, it runs the sin(1) + cos(2)
following commands automatically: >>> expr.subs({x:1, y:2}).n()
>>> from __future__ import division 0.425324148260754
>>> from sympy import *
>>> x, y, z, t = symbols('x y z t')
Note how we used .n() to obtain the expression’s numeric value.
>>> k, m, n = symbols('k m n', integer=True)
>>> f, g, h = symbols('f g h', cls=Function)
Solving equations
The first statement instructs python to convert 1/7 to 1.0/7 when >>> from sympy import solve
dividing, potentially saving you from any int division confusion. The
second statement imports all the SymPy functions. The remaining The function solve is the main workhorse in SymPy. This incredibly
statements define some generic symbols x, y, z, and t, and several powerful function knows how to solve all kinds of equations. In fact
other symbols with special properties. solve can solve pretty much any equation! When high school students
learn about this function, they get really angry—why did they spend
Note the difference between the following two statements: five years of their life learning to solve various equations by hand,
>>> x + 2 when all along there was this solve thing that could do all the math
x + 2 # an Add expression for them? Don’t worry, learning math is never a waste of time.
>>> p + 2
NameError: name 'p' is not defined The function solve takes two arguments. Use solve(expr,var) to
solve the equation expr==0 for the variable var. You can rewrite any
The name x is defined as a symbol, so SymPy knows that x + 2 is an equation in the form expr==0 by moving all the terms to one side
expression; but the variable p is not defined, so SymPy doesn’t know of the equation; the solutions to A(x) = B(x) are the same as the
what to make of p + 2. To use p in expressions, you must first define solutions to A(x) − B(x) = 0.
it as a symbol:
>>> p = Symbol('p') # the same as p = symbols('p') For example, to solve the quadratic equation x2 + 2x − 8 = 0, use
>>> p + 2 >>> solve( x**2 + 2*x - 8, x)
 3

[2, -4] Exponentials and logarithms

In this case the equation has two solutions so solve returns a list. Euler’s number e = 2.71828 . . . is defined one of several ways,
Check that x = 2 and x = −4 satisfy the equation x2 + 2x − 8 = 0. ∞
1 1
 n X
e ≡ lim 1+ ≡ lim (1 + )1/ ≡ ,
The best part about solve and SymPy is that you can obtain symbolic n→∞ n →0 n!
answers when solving equations. Instead of solving one specific n=0

quadratic equation, we can solve all possible equations of the form and is denoted E in SymPy. Using exp(x) is equivalent to E**x.
ax2 + bx + c = 0 using the following steps: The functions log and ln both compute the logarithm base e:
>>> a, b, c = symbols('a b c') >>> log(E**3) # same as ln(E**3)
>>> solve( a*x**2 + b*x + c, x) 3
[(-b + sqrt(b**2 - 4*a*c))/(2*a), (-b-sqrt(b**2-4*a*c))/(2*a)]
By default, SymPy assumes the inputs to functions like exp and log are
In this case solve calculated the solution in terms of the symbols complex numbers, so it will not expand certain logarithmic expressions.
a, b, and c. You should be able to recognize the
√ expressions in the However, indicating to SymPy that the inputs are positive real numbers
solution—it’s the quadratic formula x1,2 =
−b± b2 −4ac
. will make the expansions work:
2a
>>> x, y = symbols('x y')
To solve a specific equation like x2 + 2x − 8 = 0, we can substitute >>> log(x*y).expand()
the coefficients a = 1, b = 2, and c = −8 into the general solution to log(x*y)
obtain the same result: >>> a, b = symbols('a b', positive=True)
>>> gen_sol = solve( a*x**2 + b*x + c, x) >>> log(a*b).expand()
>>> [ gen_sol[0].subs({'a':1,'b':2,'c':-8}), log(a) + log(b)
gen_sol[1].subs({'a':1,'b':2,'c':-8}) ]
[2, -4]
Polynomials

Let’s define a polynomial P with roots at x = 1, x = 2, and x = 3:

To solve a system of equations, you can feed solve with the list of >>> P = (x-1)*(x-2)*(x-3)
equations as the first argument, and specify the list of unknowns you >>> P
want to solve for as the second argument. For example, to solve for x (x - 1)*(x - 2)*(x - 3)
and y in the system of equations x + y = 3 and 3x − 2y = 0, use
To see the expanded version of the polynomial, call its expand
>>> solve([x + y - 3, 3*x - 2*y], [x, y])
method:
{x: 6/5, y: 9/5}
>>> P.expand()
x**3 - 6*x**2 + 11*x - 6
The function solve is like a Swiss Army knife you can use to solve When the polynomial is expressed in it’s expanded form P (x) =
all kind of problems. Suppose you want to complete the square in the x3 − 62 + 11x − 6, we can’t immediately identify its roots. This is
expression x2 − 4x + 7, that is, you want to find constants h and k why the factored form P (x) = (x − 1)(x − 2)(x − 3) is preferable. To
such that x2 − 4x + 7 = (x − h)2 + k. There is no special “complete the factor a polynomial, call its factor method or simplify it:
square” function in SymPy, but you can call solve on the equation >>> P.factor()
(x − h)2 + k − (x2 − 4x + 7) = 0 to find the unknowns h and k: (x - 1)*(x - 2)*(x - 3)
>>> h, k = symbols('h k') >>> P.simplify()
>>> solve( (x-h)**2 + k - (x**2-4*x+7), [h,k] ) (x - 1)*(x - 2)*(x - 3)
[(2, 3)] # so h = 2 and k = 3 Recall that the roots of the polynomial P (x) are defined as the
>>> ((x-2)**2+3).expand() # verify... solutions to the equation P (x) = 0. We can use the solve function
x**2 - 4*x + 7
to find the roots of the polynomial:
Learn the basic SymPy commands and you’ll never need to suffer >>> roots = solve(P,x)
another tedious arithmetic calculation painstakingly performed by >>> roots
hand again! [1, 2, 3]
# let's check if P equals (x-1)(x-2)(x-3)
>>> simplify( P - (x-roots[0])*(x-roots[1])*(x-roots[2]) )
0
Rational functions

>>> from sympy import together, apart Equality checking

By default, SymPy will not combine or split rational expressions. In the last example, we used the simplify function to check whether
You need to use together to symbolically calculate the addition of two expressions were equal. This way of checking equality works
fractions: because P = Q if and only if P − Q = 0. This is the best way to
>>> a, b, c, d = symbols('a b c d') check if two expressions are equal in SymPy because it attempts all
>>> a/b + c/d possible simplifications when comparing the expressions. Below is
a/b + c/d a list of other ways to check whether two quantities are equal with
>>> together(a/b + c/d) example cases where they fail:
(a*d + b*c)/(b*d)
>>> p = (x-5)*(x+5)
Alternately, if you have a rational expression and want to divide the >>> q = x**2 - 25
>>> p == q # fail
numerator by the denominator, use the apart function:
False
>>> apart( (x**2+x+4)/(x+2) ) >>> p - q == 0 # fail
x - 1 + 6/(x + 2) False
 4

>>> simplify(p - q) == 0 sinh(x)

True
>>> sin(x)**2 + cos(x)**2 == 1 # fail Recall that x = cosh(µ) and y = sinh(µ) are defined as x and y
False coordinates of a point on the the hyperbola with equation x2 − y 2 = 1
>>> simplify( sin(x)**2 + cos(x)**2 - 1 ) == 0 and therefore satisfy the identity cosh2 x − sinh2 x = 1:
True >>> simplify( cosh(x)**2 - sinh(x)**2 )
1

Trigonometry
II. Complex numbers
from sympy import sin, cos, tan, trigsimp, expand_trig
>>> from sympy import I, re, im, Abs, arg, conjugate
The trigonometric functions sin and cos take inputs in radians:
>>> sin(pi/6) Ever since Newton, the word “number” has been used to refer to one
1/2 of the following types of math objects: the naturals N, the integers
>>> cos(pi/6) Z, the rationals Q, and the real numbers R. Each set of numbers is
sqrt(3)/2 associated with a different class of equations. The natural numbers
N appear as solutions of the equation m + n = x, where m and n are
For angles in degrees, you need a conversion factor of π
[rad/◦ ]:
180 natural numbers (denoted m, n ∈ N). The integers Z are the solutions
>>> sin(30*pi/180) # 30 deg = pi/6 rads to equations of the form x + m = n, where m, n ∈ N. The rational
1/2
numbers Q are necessary to solve for x in mx = n, with m, n ∈ Z.
The inverse trigonometric functions sin−1 (x) ≡ arcsin(x) and The solutions to x2 = 2 are irrational (so ∈/ Q) so we need an even
cos−1 (x) ≡ arccos(x) are used as follows: larger set that contains all possible numbers: real set of numbers R.
>>> asin(1/2) A pattern emerges where more complicated equations require the
pi/6 invention of new types of numbers.
>>> acos(sqrt(3)/2)
Consider the quadratic equation x2 = −1. There are no real solutions
pi/6 √
to this equation, but we can define an imaginary number i = −1
Recall that tan(x) ≡ cos(x)
sin(x)
. The inverse function of tan(x) is (denoted I in SymPy) that satisfies this equation:
tan (x) ≡ arctan(x) ≡ atan(x)
−1
>>> I*I
>>> tan(pi/6) -1
1/sqrt(3) # = ( 1/2 )/( sqrt(3)/2 ) >>> solve( x**2 + 1 , x)
>>> atan( 1/sqrt(3) ) [I, -I]
pi/6
The solutions are x = i and x = −i, and indeed we can verify that
The function acos returns angles in the range [0, π], while asin and i2 + 1 = 0 and (−i)2 + 1 = 0 since i2 = −1.
atan return angles in the range [− π2 , π2 ].
The complex numbers C are defined as {a + bi | a, b ∈ R}. Complex
Here are some trigonometric identities that SymPy knows: numbers contain a real part and an imaginary part:
>>> sin(x) == cos(x - pi/2) >>> z = 4 + 3*I
True >>> z
>>> simplify( sin(x)*cos(y)+cos(x)*sin(y) ) 4 + 3*I
sin(x + y) >>> re(z)
>>> e = 2*sin(x)**2 + 2*cos(x)**2 4
>>> trigsimp(e) >>> im(z)
2 3
>>> trigsimp(log(e))
log(2*sin(x)**2 + 2*cos(x)**2) The polar representation of a complex number is z ≡ |z|∠θ √ ≡ |z|e .
iθ

>>> trigsimp(log(e), deep=True) For a complex number z = a + bi, the quantity |z| = a + b is 2 2

log(2) known as the absolute value of z, and θ is its phase or its argument:
>>> simplify(sin(x)**4 - 2*cos(x)**2*sin(x)**2 + cos(x)**4) >>> Abs(z)
cos(4*x)/2 + 1/2 5
>>> arg(z)
The function trigsimp does essentially the same job as simplify. atan(3/4)
If instead of simplifying you want to expand a trig expression, you The complex conjugate of z = a + bi is the number z̄ = a − bi:
should use expand_trig, because the default expand won’t touch trig >>> conjugate( z )
functions: 4 - 3*I
>>> expand(sin(2*x)) # = (sin(2*x)).expand()
sin(2*x) Complex conjugation
√ is important for computing the absolute value
>>> expand_trig(sin(2*x)) # = (sin(2*x)).expand(trig=True) of z (|z| ≡ z z̄) and for division by z ( z1 ≡ |z|z̄ 2 ).
2*sin(x)*cos(x)

Euler’s formula
Hyperbolic trigonometric functions >>> from sympy import expand, rewrite

The hyperbolic sine and cosine in SymPy are denoted sinh and cosh Euler’s formula shows an important relation between the exponential
respectively and SymPy is smart enough to recognize them when function ex and the trigonometric functions sin(x) and cos(x):
simplifying expressions:
eix = cos x + i sin x.
>>> simplify( (exp(x)+exp(-x))/2 )
cosh(x) To obtain this result in SymPy, you must specify that the number x is
>>> simplify( (exp(x)-exp(-x))/2 ) real and also tell expand that you’re interested in complex expansions:
 5

>>> x = symbols('x', real=True) zero: limx→∞ x1 = 0. On the other hand, when x takes on smaller
>>> exp(I*x).expand(complex=True) and smaller positive values, the expression x1 becomes infinite:
cos(x) + I*sin(x) limx→0+ x1 = ∞. When x approaches 0 from the left, we have
>>> re( exp(I*x) ) limx→0− x1 = −∞. If these calculations are not clear to you, study
cos(x) the graph of f (x) = x1 .
>>> im( exp(I*x) )
sin(x) Here are some other examples of limits:
Basically, cos(x) is the real part of e , and sin(x) is the imaginary
ix >>> limit(sin(x)/x, x, 0)
part of eix . Whaaat? I know it’s weird, but weird things are bound 1
>>> limit(sin(x)**2/x, x, 0)
to happen when you input imaginary numbers to functions.
0
Euler’s formula is often used to rewrite the functions sin and cos in >>> limit(exp(x)/x**100,x,oo) # which is bigger e^x or x^100 ?
terms of complex exponentials. For example, oo # exp f >> all poly f for big x
>>> (cos(x)).rewrite(exp) Limits are used to define the derivative and the integral operations.
exp(I*x)/2 + exp(-I*x)/2
Compare this expression with the definition of hyperbolic cosine.
Derivatives

III. Calculus The derivative function, denoted f 0 (x), dx

d
f (x), dx
df
, or dx
dy
, describes
the rate of change of the function f (x). The SymPy function diff
Calculus is the study of the properties of functions. The operations of computes the derivative of any expression:
calculus are used to describe the limit behaviour of functions, calculate >>> diff(x**3, x)
their rates of change, and calculate the areas under their graphs. In 3*x**2
this section we’ll learn about the SymPy functions for calculating
The differentiation operation knows about the product rule
limits, derivatives, integrals, and summations.
[f (x)g(x)]0 = f 0 (x)g(x) + f (x)g 0 (x),
h the
i chain rule f (g(x)) =
0
0
f 0 (x)g(x)−f (x)g 0 (x)
f 0 (g(x))g 0 (x), and the quotient rule f (x)
g(x)
= g(x)2
:
Infinity
>>> diff( x**2*sin(x), x )
from sympy import oo 2*x*sin(x) + x**2*cos(x)
>>> diff( sin(x**2), x )
The infinity symbol is denoted oo (two lowercase os) in SymPy. Infinity cos(x**2)*2*x
is not a number but a process: the process of counting forever. Thus, >>> diff( x**2/sin(x), x )
∞ + 1 = ∞, ∞ is greater than any finite number, and 1/∞ is an (2*x*sin(x) - x**2*cos(x))/sin(x)**2
infinitely small number. Sympy knows how to correctly treat infinity
in expressions: The second derivative of a function f is diff(f,x,2):
>>> oo+1 >>> diff(x**3, x, 2) # same as diff(diff(x**3, x), x)
oo 6*x
>>> 5000 < oo The exponential function f (x) = ex is special because it is equal to
True its derivative:
>>> 1/oo
0 >>> diff( exp(x), x) # same as diff( E**x, x)
exp(x) # same as E**x

A differential equation is an equation that relates some unknown

Limits
function f (x) to its derivative. An example of a differential equation
from sympy import limit is f 0 (x) = f (x). What is the function f (x) which is equal to its
derivative? You can either try to guess what f (x) is or use the dsolve
We use limits to describe, with mathematical precision, infinitely large
function:
quantities, infinitely small quantities, and procedures with infinitely
many steps. >>> x = symbols('x')
>>> f = symbols('f', cls=Function) # can now use f(x)
1 n
 
The number e is defined as the limit e ≡ lim 1 + : >>> dsolve( f(x) - diff(f(x),x), f(x) )
n→∞ n f(x) == C1*exp(x)
>>> limit( (1+1/n)**n, n, oo)
E # = 2.71828182845905 We’ll discuss dsolve again in the section on mechanics.

This limit expression describes the annual growth rate of a loan with
a nominal interest rate of 100% and infinitely frequent compounding. Tangent lines
Borrow $1000 in such a scheme, and you’ll owe $2718.28 after one year.
The tangent line to the function f (x) at x = x0 is the line that passes
Limits are also useful to describe the behaviour of functions. Consider through the point (x0 , f (x0 )) and has the same slope as the function
the function f (x) = x1 . The limit command shows us what happens at that point. The tangent line to the function f (x) at the point
to f (x) near x = 0 and as x goes to infinity: x = x0 is described by the equation
>>> limit( 1/x, x, 0, dir="+")
oo T1 (x) = f (x0 ) + f 0 (x0 )(x − x0 ).
>>> limit( 1/x, x, 0, dir="-")
What is the equation of the tangent line to f (x) = 12 x2 at x0 = 1?
-oo
>>> limit( 1/x, x, oo) >>> f = S('1/2')*x**2
0 >>> f
x**2/2
As x becomes larger and larger, the fraction x1 becomes smaller >>> df = diff(f, x)
and smaller. In the limit where x goes to infinity, x1 approaches >>> df
 6

x The area under f (x) between x = a and x = b is obtained by

>>> T_1 = f.subs({x:1}) + df.subs({x:1})*(x - 1) calculating the change in the integral function:
>>> T_1 Z b
x - 1/2 # y = x - 1/2 A(a, b) = f (x) dx = F (b) − F (a).
a
The tangent line T1 (x) has the same value and slope as the function
f (x) at x = 1: In SymPy we use integrate(f, x) Rto obtain the integral function
x
>>> T_1.subs({x:1}) == f.subs({x:1}) F (x) of any function f (x): F (x) = 0 f (u) du.
True >>> integrate(x**3, x)
>>> diff(T_1, x).subs({x:1}) == diff(f, x).subs({x:1}) x**4/4
True >>> integrate(sin(x), x)
-cos(x)
>>> integrate(ln(x), x)
Optimization x*log(x) - x

This is known as an indefinite integral since the limits of integration

Optimization is about choosing an input for a function f (x) that are not defined.
results in the best value for f (x). The best value usually means the
maximum value (if the function represents something desirable like In contrast, a definite integral computes the area under f (x) between
profits) or the minimum value (if the function represents something x = a and x = b. Use integrate(f,R (x,a,b)) to compute the
b
undesirable like costs). definite integrals of the form A(a, b) = a f (x) dx:
>>> integrate(x**3, (x,0,1))
The derivative f 0 (x) encodes the information about the slope of f (x). 1/4 # the area under x^3 from x=0 to x=1
Positive slope f 0 (x) > 0 means f (x) is increasing, negative slope
f 0 (x) < 0 means f (x) is decreasing, and zero slope f 0 (x) = 0 means We can obtain
Rc the same area by first calculating
b the indefinite integral
the graph of the function is horizontal. The critical points of a function F (c) = 0 f (x) dx, then using A(a, b) = F (x)a ≡ F (b) − F (a):
f (x) are the solutions to the equation f 0 (x) = 0. Each critical point >>> F = integrate(x**3, x)
is a candidate to be either a maximum or a minimum of the function. >>> F.subs({x:1}) - F.subs({x:0})
1/4
The second derivative f 00 (x) encodes the information about the
Integrals correspond to signed area calculations:
curvature of f (x). Positive curvature means the function looks like x2 ,
negative curvature means the function looks like −x2 . >>> integrate(sin(x), (x,0,pi))
2
Let’s find the critical points of the function f (x) = x3 − 2x2 + x and >>> integrate(sin(x), (x,pi,2*pi))
use the information from its second derivative to find the maximum -2
>>> integrate(sin(x), (x,0,2*pi))
of the function on the interval x ∈ [0, 1].
0
>>> x = Symbol('x')
>>> f = x**3-2*x**2+x During the first half of its 2π-cycle, the graph of sin(x) is above the
>>> diff(f, x) x-axis, so it has a positive contribution to the area under the curve.
3*x**2 - 4*x + 1 During the second half of its cycle (from x = π to x = 2π), sin(x) is
>>> sols = solve( diff(f,x), x) below the x-axis, so it contributes negative area. Draw a graph of
>>> sols sin(x) to see what is going on.
[1/3, 1]
>>> diff(diff(f,x), x).subs( {x:sols[0]} )
-2 Fundamental theorem of calculus
>>> diff(diff(f,x), x).subs( {x:sols[1]} )
2 The integral is the “inverse operation” of the derivative. If you perform
It will help to look at the graph of this function. The point x = 1 the integral operation followed by the derivative operation on some
is a local maximum because it is a critical point of f (x) where the
3
function, you’ll obtain the same function:
curvature is negative, meaning f (x) looks like the peak of a mountain x
 Z  Z
d d
at x =
13 . The maximum value of f (x) on the interval x ∈ [0, 1] ◦ dx f (x) ≡ f (u) du = f (x).
dx dx c
is f 13 = 27 4
. The point x = 1 is a local minimum because it is a
critical point with positive curvature, meaning f (x) looks like the >>> f = x**2
bottom of a valley at x = 1. >>> F = integrate(f, x)
>>> F
x**3/3 # + C
>>> diff(F, x)
Integrals x**2

Alternately, if you compute the derivative of a function followed

The integral of f (x) corresponds to the computation of the area under by the integral, you will obtain the original function f (x) (up to a
the graph of f (x). The area under f (x) between the points x = a constant):
and x = b is denoted as follows: Z  Z x
b d
f (x) ≡ f 0 (u) du = f (x) + C.
Z
dx ◦
A(a, b) = f (x) dx. dx c
a

The integral function F corresponds to the area calculation as a >>> f = x**2

>>> df = diff(f, x)
function of the upper limit of integration:
>>> df
c
2*x
Z
F (c) ≡ f (x) dx . >>> integrate(df, x)
0
 7

x**2 # + C We sayP the series an diverges to infinity (or is divergent) while the
P
series bn converges (or is convergent). As we sum together more
The fundamental theorem of calculus is important because it tells us
and more terms of the sequence bn , the total becomes closer P∞ and
how to solve differential equations. If we have to solve for f (x) in the
closer to some finite number. In this case, the infinite sum 1
differential equation dx
d
f (x) = g(x), we can take theR integral on both n=0 n!
converges to the number e = 2.71828 . . ..
sides of the equation to obtain the answer f (x) = g(x) dx + C.
The summation command is useful because it allows us to compute
infinite sums, but for most practical applications we don’t need to take
Sequences an infinite number of terms in a series to obtain a good approximation.
This is why series are so neat: they represent a great way to obtain
Sequences are functions that take whole numbers as inputs. Instead
approximations.
of continuous inputs x ∈ R, sequences take natural numbers n ∈ N
as inputs. We denote sequences as an instead of the usual function Using standard Python commands, we can obtain an approximation
notation a(n). to e that is accurate to six decimals by summing 10 terms in the
series:
We define a sequence by specifying an expression for its nth term:
>>> import math
>>> a_n = 1/n
>>> def b_nf(n):
>>> b_n = 1/factorial(n)
return 1.0/math.factorial(n)
Substitute the desired value of n to see the value of the nth term: >>> sum( [b_nf(n) for n in range(0,10)] )
2.718281 52557319
>>> a_n.subs({n:5})
>>> E.evalf()
1/5
2.718281 82845905 # true value
The Python list comprehension syntax [item for item in list]
can be used to print the sequence values for some range of indices:
Taylor series
>>> [ a_n.subs({n:i}) for i in range(0,8) ]
[oo, 1, 1/2, 1/3, 1/4, 1/5, 1/6, 1/7]
Wait, there’s more! Not only can we use series to approximate
>>> [ b_n.subs({n:i}) for i in range(0,8) ]
[1, 1, 1/2, 1/6, 1/24, 1/120, 1/720, 1/5040] numbers, we can also use them to approximate functions.

Observe that an is not properly defined for n = 0 since 10 is a division- A power series is a series whose terms contain different powers of the
by-zero error. To be precise, we should say an ’s domain is the positive variable x. The nth term in a power series is a function of both the
naturals an : N+ → R. Observe how quickly the factorial function sequence index n and the input variable x.
n! = 1 · 2 · 3 · · · (n − 1) · n grows: 7! = 5040, 10! = 3628800, 20! > 1018 . For example, the power series of the function exp(x) = ex is
We’re often interested in calculating the limits of sequences as n → ∞. x2 x3 x4 x5 X xn ∞

What happens to the terms in the sequence when n becomes large? exp(x) ≡ 1 + x + + + + + ··· = .
2 3! 4! 5! n!
>>> limit(a_n, n, oo) n=0

0 This is, IMHO, one of the most important ideas in calculus: you can
>>> limit(b_n, n, oo) compute the value of exp(5) by taking the infinite sum of the terms
0 in the power series with x = 5:
Both an = 1
and bn = 1
converge to 0 as n → ∞. >>> exp_xn = x**n/factorial(n)
n n!
>>> summation( exp_xn.subs({x:5}), [n, 0, oo] ).evalf()
Many important math quantities are defined as limit expressions. An 148.413159102577
>>> exp(5).evalf()
interesting example to consider is the number π, which is defined as
148.413159102577 # the true value
the area of a circle of radius 1. We can approximate the area of the
unit circle by drawing a many-sided regular polygon around the circle. Note that SymPy is actually smart enough to recognize that the infinite
Splitting the n-sided regular polygon into identical triangular splices, series you’re computing corresponds to the closed-form expression e5 :
we can obtain a formula for its area An . In the limit as n → ∞, the >>> summation( exp_xn.subs({x:5}), [n, 0, oo])
n-sided-polygon approximation to the area of the unit-circle becomes exp(5)
exact:
Taking as few as 35 terms in the series is sufficient to obtain an
>>> A_n = n*tan(2*pi/(2*n))
approximation to e that is accurate to 16 decimals:
>>> limit(A_n, n, oo)
pi >>> import math # redo using only python
>>> def exp_xnf(x,n):
return x**n/math.factorial(n)
Series >>> sum( [exp_xnf(5.0,i) for i in range(0,35)] )
148.413159102577
Suppose we’re given a sequence a n and we want to compute the sum
∞ The coefficients in the power series of a function (also known as the
of all the values in this sequence n an . Series are sums of sequences.
P
Taylor series) depend on the value of the higher derivatives of the
Summing the values of a sequence an : N → R is analogous to taking
function. The formula for the nth term in the Taylor series of f (x)
the integral of a function f : R → R. (n)
expanded at x = c is an (x) = f n!(c) (x − c)n , where f (n) (c) is the
To work with series in SymPy, use the summation function whose value of the nth derivative of f (x) evaluated at x = c. The term
syntax is analogous to the integrate function: Maclaurin series refers to Taylor series expansions at x = 0.
>>> a_n = 1/n
The SymPy function series is a convenient way to obtain the series
>>> b_n = 1/factorial(n)
>>> summation(a_n, [n, 1, oo]) of any function. Calling series(expr,var,at,nmax) will show you
oo the series expansion of expr near var=at up to power nmax:
>>> summation(b_n, [n, 0, oo]) >>> series( sin(x), x, 0, 8)
E x - x**3/6 + x**5/120 - x**7/5040 + O(x**8)
 8

>>> series( cos(x), x, 0, 8) >>> u.dot(v)

1 - x**2/2 + x**4/24 - x**6/720 + O(x**8) 13
>>> series( sinh(x), x, 0, 8)
x + x**3/6 + x**5/120 + x**7/5040 + O(x**8) We can combine the algebraic and geometric formulas for the dot
>>> series( cosh(x), x, 0, 8) product to obtain the cosine of the angle between the vectors
1 + x**2/2 + x**4/24 + x**6/720 + O(x**8) u · ~v
~ ux vx + uy vy + uz vz
cos(ϕ) = = ,
Some functions are not defined at x = 0, so we expand them at a k~
ukk~v k k~
ukk~v k
different value of x. For example, the power series of ln(x) expanded and use the acos function to find the angle measure:
at x = 1 is >>> acos(u.dot(v)/(u.norm()*v.norm())).evalf()
>>> series(ln(x), x, 1, 6) # Taylor series of ln(x) at x=1 0.921263115666387 # in radians = 52.76 degrees
x - x**2/2 + x**3/3 - x**4/4 + x**5/5 + O(x**6)
Just by looking at the coordinates of the vectors ~
u and ~v , it’s difficult
Here, the result SymPy returns is misleading. The Taylor series of to determine their relative direction. Thanks to the dot product,
ln(x) expanded at x = 1 has terms of the form (x − 1)n : however, we know the angle between the vectors is 52.76◦ , which
means they kind of point in the same direction. Vectors that are at
(x − 1)2 (x − 1)3 (x − 1)4 (x − 1)5
ln(x) = (x − 1) − + − + + ··· . an angle ϕ = 90◦ are called orthogonal, meaning at right angles with
2 3 4 5 each other. The dot product of vectors for which ϕ > 90◦ is negative
Verify this is the correct formula by substituting x = 1. SymPy returns because they point mostly in opposite directions.
an answer in terms of coordinates relative to x = 1.
The notion of the “angle between vectors” applies more generally
Instead of expanding ln(x) around x = 1, we can obtain an equivalent to vectors with any numberP of dimensions. The dot product for n-
expression if we expand ln(x + 1) around x = 0: dimensional vectors is ~
n
u ·~v = i=1 ui vi . This means we can talk about
>>> series(ln(x+1), x, 0, 6) # Maclaurin series of ln(x+1) “the angle between” 1000-dimensional vectors. That’s pretty crazy
x - x**2/2 + x**3/3 - x**4/4 + x**5/5 + O(x**6) if you think about it—there is no way we could possibly “visualize”
1000-dimensional vectors, yet given two such vectors we can tell if
they point mostly in the same direction, in perpendicular directions,
IV. Vectors
or mostly in opposite directions.
A vector ~v ∈ Rn is an n-tuple of real numbers. For example, consider The dot product is a commutative operation ~
u · ~v = ~v · ~
u:
a vector that has three components: >>> u.dot(v) == v.dot(u)
True
~v = (v1 , v2 , v3 ) ∈ (R, R, R) ≡ R3 .
Projections
To specify the vector ~v , we specify the values for its three components
v1 , v2 , and v3 . Dot products are used for computing projections. Assume you’re
given two vectors ~
u and ~n and you want to find the component of ~u
A matrix A ∈ Rm×n is a rectangular array of real numbers with m that points in the ~n direction. The following formula based on the
rows and n columns. A vector is a special type of matrix; we can dot product will give you the answer:
think of a vector ~v ∈ Rn either as a row vector (1 × n matrix) or a
u · ~n
~
column vector (n × 1 matrix). Because of this equivalence between Π~n (~
u) ≡ ~n.
k~nk2
vectors and matrices, there is no need for a special vector object in
SymPy, and Matrix objects are used for vectors as well. This is how to implement this formula in SymPy:
>>> u = Matrix([4,5,6])
This is how we define vectors and compute their properties: >>> n = Matrix([1,1,1])
>>> u = Matrix([[4,5,6]]) # a row vector = 1x3 matrix >>> (u.dot(n) / n.norm()**2)*n
>>> v = Matrix([[7], [5, 5, 5] # projection of v in the n dir
[8], # a col vector = 3x1 matrix
[9]]) In the case where the direction vector n̂ is of unit length kn̂k = 1,
>>> v.T # use the transpose operation to the projection formula simplifies to Πn̂ (~
u) ≡ (~
u · n̂)n̂.
Matrix([[7, 8, 9]]) # convert a col vec to a row vec
Consider now the plane P defined by (1, 1, 1) · [(x, y, z) − (0, 0, 0)] = 0.
A plane is a two dimensional subspace of R3 . We can decompose any
>>> u[0] # 0-based indexing for entries
4 vector ~
u ∈ R3 into two parts ~u = ~v + w
~ such that ~v lies inside the
>>> u.norm() # length of u plane and w~ is perpendicular to the plane (parallel to ~n = (1, 1, 1)).
sqrt(77) To obtain the perpendicular-to-P component of ~
u, compute the
>>> uhat = u/u.norm() # unit vector in same dir as u projection of ~
u in the direction ~n:
>>> uhat
>>> w = (u.dot(n) / n.norm()**2)*n
[4/sqrt(77), 5/sqrt(77), 6/sqrt(77)]
[5, 5, 5]
>>> uhat.norm()
1 To obtain the in-the-plane-P component of ~
u, start with ~
u and
subtract the perpendicular-to-P part:
>>> v = u - (u.dot(n)/n.norm()**2)*n # same as u - w
Dot product [ -1, 0, 1]

The dot product of the 3-vectors ~

u and ~v can be defined two ways: You should check on your own that ~v + w
~ =~
u as claimed.
Cross product
u · ~v ≡ ux vx + uy vy + uz vz ≡ k~
~ ukk~v k cos(ϕ) ∈ R,
The cross product, denoted ×, takes two vectors as inputs and
| {z } | {z }
algebraic def. geometric def.
produces a vector as output. The cross products of individual basis
where ϕ is the angle between the vectors ~
u and ~v . In SymPy,
elements are defined as follows:
>>> u = Matrix([ 4,5,6])
>>> v = Matrix([-1,1,2]) ı̂ × ̂ = k̂, ̂ × k̂ = ı̂, k̂ × ı̂ = ̂.
 9

Here is how to compute the cross product of two vectors in SymPy: Kinematics
>>> u = Matrix([ 4,5,6])
>>> v = Matrix([-1,1,2]) Let x(t) denote the position of an object, v(t) denote its velocity, and
>>> u.cross(v) a(t) denote its acceleration. Together x(t), v(t), and a(t) are known
[4, -14, 9] as the equations of motion of the object.
The vector ~u × ~v is orthogonal to both ~ u and ~v . The norm of the
cross product k~u × ~v k is proportional to the lengths of the vectors The equations of motion are related by the derivative operation:
and the sine of the angle between them: d d
dt dt
(u.cross(v).norm()/(u.norm()*v.norm())).n() a(t) ←− v(t) ←− x(t).
0.796366206088088 # = sin(0.921..) Assume we know the initial position xi ≡ x(0) and the initial velocity
The name “cross product” is well-suited for this operation since it is vi ≡ v(0) of the object and we want to find x(t) for all later times.
calculated by “cross-multiplying” the coefficients of the vectors: We can do this starting from the dynamics of the problem—the forces
acting on the object.
u × ~v = (uy vz − uz vy , uz vx − ux vz , ux vy − uy vx ) .
~
Newton’s second law F ~net = m~a states that a net force F
~net applied
By defining individual symbols for the entries of two vectors, we can on an object of mass m produces acceleration ~a. Thus, we can obtain
make SymPy show us the cross-product formula: an objects acceleration if we know the net force acting on it. Starting
>>> u1,u2,u3 = symbols('u1:4') from the knowledge of a(t), we can obtain v(t) by integrating then
>>> v1,v2,v3 = symbols('v1:4') find x(t) by integrating v(t):
>>> Matrix([u1,u2,u3]).cross(Matrix([v1,v2,v3])) R R
vi + dt xi + dt
[ (u2*v3 - u3*v2), (-u1*v3 + u3*v1), (u1*v2 - u2*v1) ] a(t) −→ v(t) −→ x(t).
The dot product is anti-commutative ~
u × ~v = −~v × ~
u: The reasoning follows from the fundamental theorem of calculus: if
>>> u.cross(v) a(t) represents the change in v(t), then the total of a(t) accumulated
[4, -14, 9] between t = t1 and t = t2 is equal to the total change in v(t) between
>>> v.cross(u) these times: ∆v = v(t2 ) − v(t1 ). Similarly, the integral of v(t) from
[-4, 14,-9]
t = 0 until t = τ is equal to x(τ ) − x(0).
The product of two numbers and the dot product of two vectors
are commutative operations. The cross product, however, is not
commutative: ~
u × ~v 6= ~v × ~
u.
Uniform acceleration motion (UAM)

V. Mechanics Let’s analyze the case where the net force on the object is constant.
A constant force causes a constant acceleration a = mF
= constant. If
The module called sympy.physics.mechanics contains elaborate the acceleration function is constant over time a(t) = a. We find v(t)
tools for describing mechanical systems, manipulating reference and x(t) as follows:
frames, forces, and torques. These specialized functions are not >>> t, a, v_i, x_i = symbols('t a v_i x_i')
necessary for a first-year mechanics course. The basic SymPy functions >>> v = v_i + integrate(a, (t, 0,t) )
like solve, and the vector operations you learned in the previous >>> v
sections are powerful enough for basic Newtonian mechanics. a*t + v_i
>>> x = x_i + integrate(v, (t, 0,t) )
>>> x
Dynamics a*t**2/2 + v_i*t + x_i

You may remember these equations from your high school physics
The net force acting on
P an object is the sum of all the external forces
class. They are the uniform accelerated motion (UAM) equations:
acting on it F~net = ~ . Since forces are vectors, we need to use
F
vector addition to compute the net force. a(t) = a,
Compute F
~net = F
~1 + F
~2 , where F
~1 = 4ı̂[N] and F
~2 = 5∠30◦ [N]: v(t) = vi + at,
1
>>> F_1 = Matrix( [4,0] ) x(t) = xi + vi t + at2 .
2
>>> F_2 = Matrix( [5*cos(30*pi/180), 5*sin(30*pi/180) ] )
In high school, you probably had to memorize these equations. Now
>>> F_net = F_1 + F_2
>>> F_net you know how to derive them yourself starting from first principles.
[4 + 5*sqrt(3)/2, 5/2] # in Newtons
For the sake of completeness, we’ll now derive the fourth UAM
>>> F_net.evalf()
equation, which relates the object’s final velocity to the initial velocity,
[8.33012701892219, 2.5] # in Newtons
the displacement, and the acceleration, without reference to time:
To express the answer in length-and-direction notation, use norm to >>> (v*v).expand()
find the length of F
~net and atan21 to find its direction:
a**2*t**2 + 2*a*t*v_i + v_i**2
>>> F_net.norm().evalf() >>> ((v*v).expand() - 2*a*x).simplify()
8.69718438067042 # |F_net| in [N] -2*a*x_i + v_i**2
>>> (atan2( F_net[1],F_net[0] )*180/pi).n()
16.7053138060100 # angle in degrees The above calculation shows vf2 − 2axf = −2axi + vi2 . After moving
the term 2axf to the other side of the equation, we obtain
The net force on the object is F
~net = 8.697∠16.7◦ [N].
(v(t))2 = vf2 = vi2 + 2a∆x = vi2 + 2a(xf − xi ).
1 The function atan2(y,x) computes the correct direction for all vectors
(x, y), unlike atan(y/x) which requires corrections for angles in the range The fourth equation is important for practical purposes because it
[ π2 , 3π
2
]. allows us to solve physics problems without using the time variable.
 10

Example: Find the position function of an object at time t = 3[s], if The motion of a mass-spring system is described by the pdifferential
d2
it starts from xi = 20[m] with vi = 10[m/s] and undergoes a constant equation dt 2 x(t) + ω x(t) = 0, where the constant ω =
2
m
k
is called
acceleration of a = 5[m/s2 ]. What is the object’s velocity at t = 3[s]? the angular frequency. We can find the position function x(t) using
>>> x_i = 20 # initial position the dsolve method:
>>> v_i = 10 # initial velocity >>> t = Symbol('t') # time t
>>> a = 5 # acceleration (constant during motion) >>> x = Function('x') # position function x(t)
>>> x = x_i + integrate( v_i+integrate(a,(t,0,t)), (t,0,t) ) >>> w = Symbol('w', positive=True) # angular frequency w
>>> x >>> sol = dsolve( diff(x(t),t,t) + w**2*x(t), x(t) )
5*t**2/2 + 10*t + 20 >>> sol
>>> x.subs({t:3}).n() # x(3) in [m] x(t) == C1*sin(w*t) + C2*cos(w*t)
72.5 >>> x = sol.rhs
>>> diff(x,t).subs({t:3}).n() # v(3) in [m/s] >>> x
25 # = sqrt( v_i**2 + 2*a*52.5 ) C1*sin(w*t) + C2*cos(w*t)
If you think about it, physics knowledge combined with computer Note the solution x(t) = C1 sin(ωt) + C2 cos(ωt) is equivalent to
skills is like a superpower! x(t) = A cos(ωt + φ), which is more commonly used to describe
simple harmonic motion. We can use the expand function with the
General equations of motion argument trig=True to convince ourselves of this equivalence:
R R >>> A, phi = symbols("A phi")
vi + dt xi + dt >>> (A*cos(w*t - phi)).expand(trig=True)
The procedure a(t) −→ v(t) −→ x(t) can be used to obtain A*sin(phi)*sin(w*t) + A*cos(phi)*cos(w*t)
the position function x(t) even when the acceleration
√ is not constant.
Suppose the acceleration of an object is a(t) = kt; what is its x(t)? If we define C1 = A sin(φ) and C2 = A cos(φ), we obtain the form
>>> t, v_i, x_i, k = symbols('t v_i x_i k') x(t) = C1 sin(ωt) + C2 cos(ωt) that SymPy found.
>>> a = sqrt(k*t) Conservation of energy: We can verify that the total energy of the
>>> x = x_i + integrate( v_i+integrate(a,(t,0,t)), (t, 0,t) )
mass-spring system is conserved by showing ET (t) = Us (t) + K(t) =
>>> x
x_i + v_i*t + (4/15)*(k*t)**(5/2)/k**2 constant:
>>> x = sol.rhs.subs({"C1":0,"C2":A})
>>> x
Potential energy A*cos(t*w)
>>> v = diff(x, t)
Instead of working with the kinematic equations of motion x(t), v(t), -A*w*sin(t*w)
and a(t) which depend on time, we can solve physics problems using >>> E_T = (0.5*k*x**2 + 0.5*m*v**2).simplify()
energy calculations. A key connection between the world of forces and >>> E_T
the world of energy is the concept of potential energy. If you move 0.5*A**2*(k*cos(w*t)**2 + m*w**2*sin(w*t)**2)
an object against a conservative force (think raising a ball in the air >>> E_T.subs({k:m*w**2}).simplify()
against the force of gravity), you can think of the work you do agains 0.5*m*(w*A)**2 # = K_max
the force as being stored in the potential energy of the object. >>> E_T.subs({w:sqrt(k/m)}).simplify()
0.5*k*A**2 # = U_max
For each force F
~ (x) there is a corresponding potential energy UF (x).
The change in potential energy associated with the force F ~ (x) and
displacement d~ is defined as the negative of the work done by the VI. Linear algebra
force during the displacement: UF (x) = −W = − d~ F ~ (x) · d~
x.
R
from sympy import Matrix

The potential energies associated with gravity F ~g = −mg̂ and the A matrix A ∈ Rm×n is a rectangular array of real numbers with m
force of a spring F
~s = −k~
x are calculated as follows: rows and n columns. To specify a matrix A, we specify the values for
>>> x, y = symbols('x y') its mn components a11 , a12 , . . . , amn as a list of lists:
>>> m, g, k, h = symbols('m g k h') >>> A = Matrix( [[ 2,-3,-8, 7],
>>> F_g = -m*g # Force of gravity on mass m [-2,-1, 2,-7],
>>> U_g = - integrate( F_g, (y,0,h) ) [ 1, 0,-3, 6]] )
>>> U_g
m*g*h # Grav. potential energy Use the square brackets to access the matrix elements or to obtain a
>>> F_s = -k*x # Spring force for displacement x submatrix:
>>> U_s = - integrate( F_s, (x,0,x) ) >>> A[0,1] # row 0, col 1of A
>>> U_s -3
k*x**2/2 # Spring potential energy >>> A[0:2,0:3] # top-left 2x3 submatrix of A
[ 2, -3, -8]
Note the negative sign in the formula defining the potential energy.
[-2, -1, 2]
This negative is canceled by the negative sign of the dot product F ·d~
~ x:
when the force acts in the direction opposite to the displacement, the Some commonly used matrices can be created with shortcut methods:
work done by the force is negative. >>> eye(2) # 2x2 identity matrix
[1, 0]
[0, 1]
Simple harmonic motion >>> zeros((2, 3))
from sympy import Function, dsolve [0, 0, 0]
[0, 0, 0]
The force exerted by a spring is given by the formula F = −kx. If
the only force acting on a mass m is the force of a spring, we can use Standard algebraic operations like addition +, subtraction -, multipli-
Newton’s second law to obtain the following equation: cation *, and exponentiation ** work as expected for Matrix objects.
The transpose operation flips the matrix through its diagonal:
d2
h i
F = ma ⇒ −kx = ma ⇒ −kx(t) = m x(t) . >>> A.transpose() # the same as A.T
dt2
 11

[ 2, -2, 1] The column space of A is the span of the columns of A that contain
[-3, -1, 0] the pivots in the reduced row echelon form of A:
[-8, 2, -3] >>> [ A[:,c] for c in A.rref()[1] ] # C(A)
[ 7, -7, 6] [ [ 2] [-3] [-8]
Recall that the transpose is also used to convert row vectors into [-2], [-1], [ 2]
column vectors and vice versa. [ 1] [ 0] [-3] ]

Note we took columns from the original matrix A and not its RREF.
Row operations To find the null space of A, call its nullspace method:
>>> M = eye(3) >>> A.nullspace() # N(A)
>>> M[1,:] = M[1,:] + 3*M[0,:] [ [0, -3, 2, 1] ]
>>> M
[1, 0, 0]
[3, 1, 0] Determinants
[0, 0, 1]
The determinant of a matrix, denoted det(A) or |A|, is a particular
The notation M[i,:] refers to entire rows of the matrix. The first way to multiply the entries of the matrix to produce a single number.
argument specifies the 0-based row index, for example the first row >>> M = Matrix( [[1, 2, 3],
of M is M[0,:]. The code example above implements the row operation [2,-2, 4],
R2 ← R2 + 3R1 . To scale a row i by constant c, use the command [2, 2, 5]] )
M[i,:] = c*M[i,:]. To swap rows i and j, use can use the Python >>> M.det()
tuple-assignment syntax M[i,:], M[j,:] = M[j,:], M[i,:]. 2

Determinants are used for all kinds of tasks: to compute areas and
Reduced row echelon form volumes, to solve systems of equations, and to check whether a matrix
The Gauss–Jordan elimination procedure is a sequence of row is invertible or not.
operations you can perform on any matrix to bring it to its reduced
row echelon form (RREF). In SymPy, matrices have a rref method
Matrix inverse
that computes their RREF:
>>> A = Matrix( [[2,-3,-8, 7], For every invertible matrix A, there exists an inverse matrix A−1
[-2,-1,2,-7], which undoes the effect of A. The cumulative effect of the product of
[1 ,0,-3, 6]]) A and A−1 (in any order) is the identity matrix: AA−1 = A−1 A = 1.
>>> A.rref()
([1, 0, 0, 0] # RREF of A >>> A = Matrix( [[1,2],
[0, 1, 0, 3] # locations of pivots [3,9]] )
[0, 0, 1, -2], [0, 1, 2] ) >>> A.inv()
[ 3, -2/3]
Note the rref method returns a tuple of values: the first value is the [-1, 1/3]
RREF of A, while the second tells you the indices of the leading ones >>> A.inv()*A
(also known as pivots) in the RREF of A. To get just the RREF of [1, 0]
A, select the 0th entry form the tuple: A.rref()[0]. [0, 1]
>>> A*A.inv()
[1, 0]
Matrix fundamental spaces [0, 1]

Consider the matrix A ∈ Rm×n . The fundamental spaces of a matrix The matrix inverse A−1 plays the role of division by A.
are its column space C(A), its null space N (A), and its row space
R(A). These vector spaces are important when you consider the Eigenvectors and eigenvalues
matrix product A~x = ~ y as “applying” the linear transformation
When a matrix is multiplied by one of its eigenvectors the output
TA : Rn → Rm to an input vector ~ x ∈ Rn to produce the output
is the same eigenvector multiplied by a constant A~eλ = λ~eλ . The
vector ~
y∈R . m
constant λ (the Greek letter lambda) is called an eigenvalue of A.
Linear transformations TA : Rn → Rm (vector functions) are
To find the eigenvalues of a matrix, start from the definition A~eλ =
equivalent to m × n matrices. This is one of the fundamental ideas
λ~eλ , insert the identity 1, and rewrite it as a null-space problem:
in linear algebra. You can think of TA as the abstract description of
the transformation and A ∈ Rm×n as a concrete implementation of A~eλ = λ1~eλ ⇒ (A − λ1) ~eλ = ~0.
TA . By this equivalence, the fundamental spaces of a matrix A tell
us facts about the domain and image of the linear transformation This equation will have a solution whenever |A − λ1| = 0.2 The
TA . The columns space C(A) is the same as the image space space eigenvalues of A ∈ Rn×n , denoted {λ1 , λ2 , . . . , λn }, are the roots of
Im(TA ) (the set of all possible outputs). The null space N (A) is the the characteristic polynomial p(λ) = |A − λ1|.
same as the kernel Ker(TA ) (the set of inputs that TA maps to the >>> A = Matrix( [[ 9, -2],
zero vector). The row space R(A) is the orthogonal complement of [-2, 6]] )
>>> A.eigenvals() # same as solve( det(A-eye(2)*x), x)
the null space. Input vectors in the row space of A are in one-to-one
{5: 1, 10: 1} # eigenvalues 5 and 10 with multiplicity 1
correspondence with the output vectors in the column space of A. >>> A.eigenvects()
Okay, enough theory! Let’s see how to compute the fundamental [(5, 1, [ 1]
spaces of the matrix A defined above. The non-zero rows in the [ 2] ), (10, 1, [-2]
[ 1] )]
reduced row echelon form of A are a basis for its row space:
>>> [ A.rref()[0][r,:] for r in A.rref()[1] ] # R(A) 2 The invertible matrix theorem states that a matrix has a non-empty
[ [1, 0, 0, 0], [0, 1, 0, 3], [0, 0, 1, -2] ] null space if and only if its determinant is zero.
 12

Certain matrices can be written entirely in terms of their eigenvectors Links

and their eigenvalues. Consider the matrix Λ (capital Greek L) that
has the eigenvalues of the matrix A on the diagonal, and the matrix [ Installation instructions for jupyter notebook ]
Q constructed from the eigenvectors of A as columns: https://jupyter.readthedocs.io/en/latest/install.html
[ The official SymPy tutorial ]
  
0
 
λ1 ···  |
Λ = . .. http://docs.sympy.org/latest/tutorial/intro.html
 λ1 · · · ~eλn , then A = QΛQ .
−1
 ..
, Q =~e
.

0
0 0 [ A list of SymPy gotchas ]

λn  |
http://docs.sympy.org/dev/gotchas.html
Matrices that can be written this way are called diagonalizable. To [ SymPy video tutorials by Matthew Rocklin ]
diagonalize a matrix A is to find its Q and Λ matrices: http://pyvideo.org/speaker/583/matthew-rocklin
>>> Q, L = A.diagonalize()
>>> Q # the matrix of eigenvectors
[1, -2] # as columns Book plug
[2, 1]
>>> Q.inv() The examples and math explanations in this tutorial are sourced from
[ 1/5, 2/5] the no bullshit guide series of books published by Minireference Co.
[-2/5, 1/5] We publish textbooks that make math and physics accessible and
>>> L # the matrix of eigenvalues affordable for everyone. If you’re interested in learning more about
[5, 0] the math, physics, and calculus topics discussed in this tutorial,
[0, 10] check out the No bullshit guide to math and physics. The book
>>> Q*L*Q.inv() # eigendecomposition of A
contains the distilled information that normally comes in two first-
[ 9, -2]
[-2, 6] year university books: the introductory physics book (1000+ pages)
>>> Q.inv()*A*Q # obtain L from A and Q and the first-year calculus book (1000+ pages). Would you believe
[5, 0] me if I told you that you can learn the same material from a single
[0, 10] book that is 1/7th the size and 1/10th of the price of mainstream
textbooks?
Not all matrices are diagonalizable. You can check if a matrix is
diagonalizable by calling its is_diagonalizable method: This book contains short lessons
>>> A.is_diagonalizable() on math and physics, written in
True a style that is jargon-free and
>>> B = Matrix( [[1, 3], to the point. Often calculus and
[0, 1]] ) mechanics are taught as sepa-
>>> B.is_diagonalizable() rate subjects. It shouldn’t be like
False that. If you learn calculus with-
>>> B.eigenvals() out mechanics, it will be boring.
{1: 2} # eigenvalue 1 with multiplicity 2
If you learn mechanics without
>>> B.eigenvects()
[(1, 2, [1] calculus, you won’t truly under-
[0] )] stand what is going on. This
textbook covers both subjects
The matrix B is not diagonalizable because it doesn’t have a full set in an integrated manner.
of eigenvectors. To diagonalize a 2 × 2 matrix, we need two orthogonal
eigenvectors but B has only a single eigenvector. Therefore, we can’t Contents:
construct the matrix of eigenvectors Q (we’re missing a column!) and
• high school math
so B is not diagonalizable.
• vectors
Non-square matrices don’t have eigenvectors and therefore don’t • mechanics
have an eigendecomposition. Instead, we can use the singular value • differential calculus
decomposition to break up a non-square matrix A into left singular • integral calculus
vectors, right singular vectors, and a diagonal matrix of singular • 250+ practice problems
values. Use the singular_values method on any matrix to find its
singular values. 5½[in] × 8½[in] × 445[pages]

For more information, see the book’s website at minireference.com.

Conclusion
The linear algebra examples presented in Section VI are sourced
I would like to conclude with some words of caution about the overuse from the No bullshit guide to linear algebra. Check out the
of computers. Computer technology is very powerful and is everywhere book if you’re taking a linear algebra course of if you’re missing the
around us, but let’s not forget that computers are actually very dumb: prerequisites for learning machine learning, computer graphics, or
computers are mere calculators and they depend on your knowledge quantum mechanics.
to direct them. It’s important that you learn how to do complicated
I’ll close on a note for potential readers who suffer from math-phobia.
math by hand in order to be able to instruct computers to do math
Both books start with an introductory chapter that reviews all high
for you and to check the results of your computer calculations. I
school math concepts needed to make math and physics accessible to
don’t want you to use the tricks you learned in this tutorial to avoid
everyone. Don’t worry, we’ll fix this math-phobia thing right up for
math problems from now on and simply rely blindly on SymPy for all
you; when you’ve got SymPy skills, math fears you!
your math needs. I want both you and the computer to become math
powerhouses! The computer will help you with tedious calculations To stay informed about upcoming titles, follow @minireference on
(they’re good at that) and you’ll help the computer by guiding it twitter and check out the facebook page at fb.me/noBSguide.
when it gets stuck (humans are good at that).