6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
6.001 Notes,Section 3.1
Slide 3.1.1
In this lecture,we are going to put together the basic pieces of
Scheme that we introduced in the previous lecture,in order to
start capturing computational processes inside procedures,To
do this,we will introduce a model of computation,called the
substitution model,We will show how that model helps us
relate the choices we make in designing a procedure to the
actual process of evaluation that will occur when we use that
procedure,
Thus we will first look at an example of the substitution model,
to make sure you understand its mechanics,and then we will
use to that examine two different approaches to creating
procedures,
Slide 3.1.2
Now that we have the ability to create procedures and use them,
we need some way of figuring out what happens during the
evolution of a procedure application,For that we have
something called the substitution model,You've actually seen
this,but we just didn't call it that,and now we are going to
make it quite explicit,
The role of the substitution model is to provide us with a means
of determining how an expression evolves,In examining the
substitution model,we stress that this is not a perfect model of
what goes on inside the computer,In fact,later in the term we
will see a much more detailed and accurate model of
computation,but this model suffices for our purposes,Indeed,
we don't really care about using the substitution model to figure out the actual value of an expression,We can use
the computer to do that,Rather,we want to use the substitution model to understand the process of evaluation,so
that we can use that understanding to reason about choices in designing procedures,Given that understanding,we
can work backwards,using a desired pattern of evaluation to help us design the correct procedure to generate that
evaluation pattern,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.1.3
So here are the rules for the substitution model,Given an
expression,we want to determine its value,If the expression is
a simple expression there are several possibilities,If it is a self-
evaluating expression,like a number,we just return that value,
If the expression is a name (something we created with a define
expression) then we replace the name with the corresponding
value associated with it,
If the expression is a special form,there are also several
possibilities,If the expression is a lambda then we replace the
expression with some representation of the actual procedure,If
the expression is some other special form (such as an if
expression) then we follow specific rules for evaluating the
subexpressions of that special form,
Otherwise the expression is a compound expression (one of those expressions nested in parentheses that contains
several subexpressions),Here we apply exactly the same set of rules to each of the subexpressions of the
compound expression,in some arbitrary order,We then look at the value of the first subexpression,If it is a
primitive procedure (one of the built-in procedures,like * or +) then we just apply that procedure to the values of
the other expressions,On the other hand,if the value of the first subexpression is a compound procedure
(something we created by evaluating a lambda expression) then we substitute the value of each subsequent
subexpression for the corresponding procedure parameter in the body of the procedure,We replace the entire
expression with this instantiated body,and repeat the process,
Slide 3.1.4
So here is a simple example,Suppose we have defined square
to be the obvious procedure that multiplies a value by itself,
Now,let's use the substitution model to trace out the evolution
of applying that procedure,
The substitution model says that to trace the evaluation of
(square 4),we first determine that this is a compound
expression,Since it is,we evaluate each of the subexpressions
using the same rules,Thus,the first subexpression,square,is
just a name,and we look up its value,getting back the
procedure we created when we evaluated the lambda,The
second subexpression is just a number,so that is its value,Now
the rule says to substitute 4 everywhere we see the formal
parameter x in the body of the procedure and replace the original expression with this instantiated body expression,
as shown,
This reduces the evaluation of (square 4) to the evaluation of the expression (* 4 4),This is also a compound
expression,so again we get the values of the subexpressions,In this case,the application is of a primitive
procedure,so we simply apply it to reduce the expression to its final value,16,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.1.5
Here is a more detailed example,First,let's create two
procedures,These expressions will each generate a procedure,
through the lambda expressions,and the defines will then
associate a name with each of them,Thus,in our environment,
we have pairings of square and average to the associated
procedure objects,each with its own parameter list and body,
Slide 3.1.6
Here is an example using these procedures,To evaluate this
first expression using the substitution model,I first need to get
the values of the sub-expressions,The value of average I get by
looking this name up in the environment,giving me the
associated procedure object,The value of 5 is easy,The last sub-
expression is itself a combination,so I recursively apply the
rules,By the same reasoning I get the procedure associated with
square and the value 3,then substitute 3 into the body of square
for x,This reduces to another combination,which I can
recognize is an application of a primitive procedure,so this
reduces to the number 9,Thus the recursive application of the
substitution model yields a simpler combination,
Note the format here,I first evaluate the sub-expressions,then substitute and apply the procedure,This is called an
applicative order evaluation model,
Slide 3.1.7
Notice in particular that under this model I need to get the value
of the operands first,which caused me to evaluate that interior
compound expression before I got around to applying average,
Once I have simple values,I can continue the substitution,In
this case,I use the body of the procedure associated with
average,now substituting 5 and 9 for x and y in that body,
Note that there is no confusion about which x I am referring to,
it's the one in the procedure associated with average,not the
one in square,
As before,I substitute into the body of this procedure,and then
continue,I recursively apply the rules to each subexpression,In
the case of a compound expression,I must reduce this to a
simpler value before I can continue with the rest of the expression,Also note that when the application involves a
primitive procedure,I just do the associated operation,eventually reducing this to the final answer,
The key idea is to see how this notion of substitution lets us trace out the evolution of an expression's evaluation,
reducing an application of a procedure to the evaluation of a simpler expression,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
6.001 Notes,Section 3.2
Slide 3.2.1
Now let's look at a little less trivial example,Suppose I want to
compute factorial of n,which is defined as the product of n by
n-1 and so on down to 1,Note that I am assuming that n is an
integer and is greater than or equal to 1,
Slide 3.2.2
So how do I create a procedure to compute factorial of n? Well
I need to think carefully here,since my choice in designing this
procedure will impact the evolution of the process when I use
it,
One way to look at factorial is to group the computation into
multiplying n by all the other multiplications,but I can
recognize that this latter computation is just factorial of n-1,
So notice what this does,it reduces a computation to a simpler
operation,multiplication,and a simpler version of the same
problem,in this case factorial of a smaller number,So I have
reduced one version of a problem to a simpler version of the
same problem,plus some simple operations,This is a very
common pattern that we are going to use a lot in designing procedures,
Slide 3.2.3
In fact,given this common pattern,we can write a procedure to
capture this idea,Here it is,The first part says to give the name
fact to the procedure created by the lambda,which has a single
argument n and a particular body,Now,what does the lambda
say to do? It has two different parts,which we need to look at
carefully,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.2.4
First of all,it has a new kind of expression in the interior,the =,
This is an example of a predicate,a procedure that takes some
arguments and returns either true or false,in this case based on
the test for numerical equality,
Slide 3.2.5
The more interesting part is the body of the lambda,and it is a
new special form called an if,Because this is a special form,it
means that evaluation does not follow the normal rules for
combinations,If's are used to control the order of evaluation,
and contain three pieces,
Slide 3.2.6
Its first subexpression we call a predicate,
Slide 3.2.7
Its second subexpression we call a consequent,.,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.2.8
..,and its third subexpression we call an alternative,Now here
is why an if is a special form,An if expression first evaluates its
predicate,using the same set of rules recursively on this
expression,It does this before it ever looks at either of the two
other expressions,Now,if the predicate evaluates to a true
value,then the if expression takes the consequent,and evaluates
it,returning that value as the value of the overall if expression,
On the other hand,if the predicate evaluates to a false value,
then the if expression takes the alternative expression,evaluates
it,and returns that value,Notice that only one of the consequent
and alternative expressions is evaluated during an if,
In the case of our fact procedure,we can now see how the
evaluation will proceed,When we apply fact to some argument,we first test to see if that value is 1,The if
expression does this by first evaluating the predicate,before ever considering the other expressions,thus changing
the order of evaluation,If the predicate is true,then we simply return the value 1,If it is false,then we will
evaluate the last expression with appropriate substitution,thus unwinding factorial of n into the multiplication of
the value of n by the result of evaluating factorial of n-1,
Slide 3.2.9
Here is a tracing of the substitution model in action,To
evaluate factorial of 3,we substitute into the body of fact,
leading to the next expression,The if first evaluates the
predicate,leading to the next expression,Since the predicate's
value is false,the entire if expression is replaced by the
alternative,with appropriate substitutions,
To evaluate this expression we need to get the values of the
subexpressions,so this entire expression reduces to a
multiplication of 3 by the value of (fact 2),Notice how this has
unwound the computation into a particular form,a deferred
multiplication and a recursive call to a simpler version of the
same problem,
This process repeats,using the if to unwind another level of the computation,and this continues,just using the
substitution model rules,until finally the predicate is true,In this case,we are left with an expression just involving
primitives,and we can complete the multiplications,thus reducing the computation to a simple answer,
Note the form shown in red,in which an expression is reduced to this pattern of a simpler operation and a simpler
version of the same problem,This unwrapping continues until we are left with a nested expression whose
innermost subexpression only involves simple expressions and operations,at which point the deferred operations
are evaluated,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.2.10
That means that fact gives rise to what we call a recursive
process,In the substitution model we can see that this is
characterized by a set of deferred operations,in which the
multiplication is deferred while we go off to get the sub-
computation of the simpler version of fact,Once we get to a
simple case,we can start accumulating those stacked up
operations,We will come back to this idea next time,
6.001 Notes,Section 3.3
Slide 3.3.1
Now that we have seen our first more interesting procedure,
fact,let's step back and generalize the ideas we used to capture
this computation in a procedure,
What are the general steps that we use in creating a recursive
procedure like this? We have three stages,as shown here,
The first stage is called wishful thinking,The idea is as
follows,If I want to create a solution to some problem,I first
assume that I have available a procedure that will solve the
problem,but only for versions of the problem smaller than the
current one,In the case of factorial,I assume (wishfully) that I
can solve factorial for problems of size smaller than n,
Slide 3.3.2
Given that assumption,the second stage proceeds to design a
solution to the problem,In particular,I can use the existence of
a solution to the smaller sized problem,plus a set of simple
operations,to design the solution to the larger sized problem,
We saw this with factorial,where we combined the solution to
a smaller version of factorial,plus a multiplication,to create the
solution to the full version of factorial,
Notice that the second step here requires some ingenuity,One
should be careful to think through this strategy,before
beginning to code up a solution,In the case of factorial,we saw
how we did this decomposition into a multiplication and a
smaller version of factorial,
With this,we can build a first version of the procedure as shown at the bottom of the slide,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.3.3
Of course you already know that this won't quite work,but let's
think about why,If we apply that version of fact to some
argument,it will keep unwinding the application of factorial
into a multiplication and a smaller version of factorial,But
before we can compute that multiplication,we have to get the
value of the smaller version of factorial,and that will continue
to unwind to another version,ad infinitum,
The problem is that I didn't use my third step; I didn't consider
the smallest sized problem that I can solve directly,without
using wishful thinking,In the case of factorial,that is just
knowing that factorial of 1 is just 1,Once I find that smallest
sized problem,I can control the unwinding of the computation,
by checking for that case and doing the direct computation when appropriate,In this way,I will terminate the
recursive unwinding of the computation and accumulate all the deferred operations,
Slide 3.3.4
This in fact is a very common form,a common pattern for a
recursive algorithm that has a test,a base case and a recursive
case,The if controls the order of evaluation to decide whether
we have the base case or the recursive case,The recursive case
controls the unwinding of the computation,doing so until we
reach the base case,
Slide 3.3.5
To summarize,we have now seen how to design recursive
algorithms,by using this idea of wishful thinking to separate
the problem into a simpler version of the same problem,do that
decomposition and identify the smallest version of the problem
that will stop the unwinding of the computation into simpler
versions of the same problem,As a consequence,algorithms
associated with this kind of approach have a test,a base case,
and a recursive case to control the order of evaluation in order
to unwind the computation into simpler versions of the same
problem,
6.001 Notes,Section 3.4
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.1
So we have introduced recursive procedures,and we have now
seen an example of building a recursive procedure to implement
factorial,Now,let's look at a different kind of procedure,an
iterative procedure for computing factorial,In doing this,we
are going to see how we can develop procedures with different
evolutions that compute the same basic computation,
Slide 3.4.2
To set the stage for this,here is our factorial procedure from
last time,Remember that this was a nice little procedure that
first checked to see if we were in the base case,and if we were,
returned the answer 1,If we were not,the procedure reduced
the problem to multiplying n by a recursive call to factorial of
n-1,One of the properties of this procedure was that it held a
set of deferred operations,To compute factorial of n,it had to
hold onto a multiplication while it went off to compute the
subproblem of a simpler version of factorial,and that in turn
required another deferred operation,Thus,the procedure had to
keep track of something to do,once it was done computing a
subproblem,and this meant that as the argument to the procedure gets larger,we would have more things to keep
track of,We would like to see if there is another way of computing factorial,without all of those deferred
operations,
Slide 3.4.3
The specific question we want to consider is whether we can
design an algorithm for factorial that uses only constant space,
By this we mean that the computation does not require keeping
track of any deferred operations,but rather can perform its
work without using up any additional memory to keep track of
things to do,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.4
So that's our goal,to compute factorial in a different way that is
not characterized by those deferred multiplications
Slide 3.4.5
..,and here is the intuition behind this different approach,Let's
think about how you might do this if you were executing the
computation by hand,One way to do this is to multiply out the
first two numbers,and keep track of that temporary product,
plus the next stage in the computation,For example,for
factorial of 4,you would multiply out 4 by 3,keep track of the
result,12,and the fact that the next thing to multiply is 2,You
would then multiply the current result,12,by 2,keeping track
of that result and the next thing to multiply by,
Slide 3.4.6
Notice that in this version,we only need to keep track of 2
things,the current product,and the next thing to do,This is
different from the recursive case,where we were blindly
keeping track of each pending multiplication,Here,we replace
the previous product by the new one,and the next multiplier by
the new one,whereas in the previous case we had to keep track
of each deferred operation,i.e,I have to multiply by 2,then I
have to multiply by 3,then,.,
Slide 3.4.7
Thus,we only need a constant amount of space,To stress,think
about the recursive case,There,we had to write down that I
have a pending operation of multiplication by some number,
plus another pending operation of multiplication by another
number,and so on,Keeping track of each such pending
operation takes a bit more space,In this case,I only need two
pieces of information,the current product and the next
multiplier,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.8
The reason I can do this is because multiplication satisfies two
nice properties from mathematics,that in essence say I can do
the multiplication in any order,and accumulating partial
answers along the way will not affect the result,Thus,I can
come up with an algorithm that only uses constant space,
Slide 3.4.9
So now let's capture that idea a little more formally,I can use
the same sort of notion in a table,in which I have one column
for each piece of information I am going to need,and one row
for each step of the computation,Each step here means how to
do I go from a current state,captured by a temporary product
and a counter that tells me the next thing to multiply,to the next
state of those parameters,
Slide 3.4.10
Now what I need to do is come up with a rule for how I am
going to change the values in the table,In particular,to get the
next value for product,I take the current value of product and
the current value of counter,multiply them,and that becomes
the next entry in the product column,
Slide 3.4.11
Similarly,I need a rule for getting the next value for counter,
That,we know,is simply given by adding 1 to the current row's
value for counter,thus generating the next row's value for
counter,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.12
Of course,I can keep repeating this process,updating the
values for the product column and the counter column until I
am ready to stop,
Slide 3.4.13
And how do I know when I am ready to stop? Well that's easy,
The last row I want in the table is the one when counter is
bigger than n,
Slide 3.4.14
Ah -- and that indicates one more thing,I will need another
column to keep track of n,since each column specifies a
needed piece of information and that is one more,The fact that
the value doesn't change is irrelevant,what matters is that I
need to keep track of this information,
Slide 3.4.15
Once I am done,I still need to get the answer out,but I know
where that is,It's sitting in the element at the intersection of the
product column and the last row,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.16
The only other thing I need to do is figure out how to start this
process off,And this is also easy,My first row will deal with
the equivalent of a base case,0! is just 1,so I can start my
product at 1,my counter at 1,set up N to be whatever value I
want to compute,and I can then start this table process
underway using the rules I just described,
Slide 3.4.17
Now let's write a procedure to do this,I am going to define a
procedure called ifact,As you can see,I use a lambda
to create the procedure of one argument,and I use a define
to give it a name,Notice that the body of the procedure is just a
call to another procedure,a helper procedure that I am also
going to define,
The reason I need a helper procedure is clear from my previous
discussion,I need to keep track of three things during the
computation,so I need a procedure with three arguments,as
shown,one for each column of my table,The body of this
helper procedure looks a lot like my earlier recursive procedure,
but with some important differences,First,there is a test case,to tell when we are done,Here we don't have a base
case,but rather there is a return value,an answer to give when the test case holds,The iterative case,the thing to
do if the test case is not true,is slightly different from our recursive case,Here,the call to the procedure has new
arguments,one for each parameter,Notice how the new argument in each case captures the update rule from the
table,
Slide 3.4.18
So now let's think about what happens if I evaluate ifact,
The value of ifact with some argument n will be the same
as evaluating the body of the procedure,that is ifact
helper with arguments 1,1,and n,Notice that this is just
setting up the initial row of my table,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.19
Then look at what happens when we evaluate ifact
helper,In fact,it just computes the next row of that table,
It says that the value of ifact-helper when called on
arguments for product,counter and n,is the same as evaluating
ifact-helper with new values for product,counter and
n,each of which exactly captures the update rules from my
table,
Slide 3.4.20
And what else do I need? Just to determine when to stop,So I
have a test to determine when that happens,and at that stage,I
know that the answer is in the product column,so I simply
return that value,
ifact on some argument,say 4,reduces to calling ifact-
Slide 3.4.21
To trace how this works,we use our substitution model,Calling
helper on arguments 1,1,and 4,since that is what the
body of ifact contains,where we have substituted 4 for
the parameter n,
Slide 3.4.22
The substitution model then says to substitute the arguments for
the formal parameters in the body of ifact-helper and
evaluate that instantiated body,namely this if expression,We
saw that if is a special form,and this means we first evaluate
the predicate,Since in this case it returns a false value,the
value of the whole if expression reduces to the value of the
alternative clause,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.23
Evaluating the expressions for the arguments then reduces to
this expression,
Notice an important thing,We have reduced the evaluation of
one expression involving ifact-helper to a different
expression involving ifact-helper,with a different set
of arguments,or parameters,But there are no deferred
operations here,that is,there are no operations that we have to
keep track of while we go off and compute some subproblem,
Rather the answer to evaluating this expression is the answer to
evaluating the first one directly,
Slide 3.4.24
And we can just repeat this process,with the same behavior,
until the predicate is true,
Notice the form of the red expressions,They are constant in
size,There are no deferred operations,no storage of
expressions while I go get the value of a subexpression,Instead,
the variables that capture the state of the computation change
with each evaluation,This is very different from the shape of
the previous version of factorial,There,we had a set of deferred
operations,and the size of the expressions grew with each step,
Here there are no deferred operations,and a constant amount of
space is needed to keep track of the computation,
Slide 3.4.25
Let's compare this to the recursive version,Remember that we
had that pending operation,and it caused the shape of the
procedure's evolution to be different,It grew with each new
evaluation,because that extra pending operation had to be
tracked,including information relevant to each operation,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.26
Compare that to the iterative version,Here there are no pending
operations,The value of one call to the procedure is the same as
the value of a different call to the same procedure,that is,
different arguments but nothing deferred,As a consequence,
this procedure has a fixed size,or constant,evolution,
So we have seen two different implementations of the same
computation,with different behaviors,
Slide 3.4.27
And this summarizes what we have seen so far,
6.001 Notes,Section 3.5
Slide 3.5.1
So how do we know that our code is actually correct? It would
be nice if we could guarantee that our code would run correctly,
provided we give it correct inputs,So how do we ensure this?
Well there are several possibilities,We could accept the word
of someone with power over our future (e.g,a 6.001 professor)
that the code is right,Clearly not a wise idea,though it is
shocking how often this approach is used! A more common
method is to rely on statistics,that is,to try our code on a bunch
of example inputs,and hope that if it works on all the ones we
try,it will work on every input,Sometimes this is the best you
can do,but clearly there are no guarantees with this approach,
All too often,programmers rely on the third method,also
known as arrogance,They just assume that they wrote correct code,and are stunned when it fails,
The best method,when possible,is to actually formally prove that our code is correct,And it is this method that we
are going to briefly explore here,We won't show you all the elements of proving programs correct,but we will
provide you with a basis for understanding how this is done,and more importantly for using the reasoning pattern
underlying this proof method to help you design good code,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.5.2
First,we need to talk about what constitutes a formal proof,
Technically,a proof of a mathematical or logical proposition is
a chain or sequence of logical deductions that starts with a base
set of axioms and leads to the proposition we are attempting to
prove,Now,we need to fill in the details on these pieces,
First,a proposition is basically a statement that is either true or
false,Typically,in propositional logic,we have a set of atomic
propositions,i.e.,simple statements of fact,For example,"n=0"
is an atomic proposition,Given some value for the variable n,
this statement is either true or false (assuming that n is a
number,of course),
More interesting propositions (and things that we are more
likely to be interested in proving) are created by combining simpler propositions,There are five standard ways of
creating compound propositions,one can take the conjunction (which acts like an "and",meaning the compound
proposition is true if and only if both of the individual propositions are true); one can take the disjunction (which
acts like an "or",meaning the compound proposition is true if and only if one or the other of the individual
propositions is true); one can take the negation of a proposition (which means the proposition is true if and only if
the initial proposition is false); we can create the implication (which means that if P is true,then Q is true); and we
can establish an equivalence (which means that P is true if and only if Q is true),
Note that compound propositions can have elements that are themselves compound propositions,so that we can
create deeply nested propositions,
Slide 3.5.3
We can capture the logical structure of these compound
propositions by simply looking at the possible cases,If we
consider all possible combinations for the base propositions,we
can chart out the veracity of the compound proposition,Thus,
we see that "P and Q" behaves as we would expect; it is true as
a statement if and only if both P and Q are true,Negation and
disjunction also behave as expected,So does equivalence,since
the combination can only be true when both simpler elements
are,
Implication is a bit more puzzling,The proposition is defined to
be true if either P is false or Q is true,To see this,take a simple
example,Suppose we consider the proposition,"If n is greater
than 2,then n squared is greater than 4",Then for example,if n equals 4,both propositions are true,and our truth
table also states that the implication is true,If for example,n equals -4,then the first proposition is false and the
second proposition is true,The compound statement is true,as we can see from the truth table,and this makes
sense,since the consequent can clearly still be true even if the precedent is not,Finally,suppose n equals 1,In this
case both propositions are false,however the statement as a whole is true,
This particular combination thus has a rather counterintuitive aspect,it is always true if the first part is false,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.5.4
Now,given that we can construct propositions,we want to be
able to prove their correctness by chaining together a sequence
of logical deductions,starting from a set of axioms (which are
propositions that are assumed to be true),The basic idea is that
we start with statements that are correct,and then use standard
deductions to reason about statements whose veracity we are
attempting to establish,
For example,a common inference rule (which has been around
at least since Aristotle) is modus ponens,This basically
says that if we have a proposition P that is true,And we have an
implication (P implies Q) that is also true,and then we can
deduce that Q is a true proposition,
A related (and equally old) inference rule is modus tollens,This basically states that if we have an
implication that P implies Q,and we also know that Q is not true,then we can infer that P is not true,
Slide 3.5.5
Now,given these tools,we could go off and prove statements
of fact,Unfortunately,the things we want to prove about
programs (e.g,does this code run for all correct inputs) require
potentially infinite propositions,since we would have to take a
conjunction of propositions,one for each input value,So we
need to extend our propositional logic to predicate logic,
We handle this by creating propositions with variables,and
then specifying the set of values (called the universe or domain
of discourse) over which the variable may range,
Since predicates apply over ranges,we also can put conditions
on those predicates,The two basic quantifiers are known as
"for all" and "there exists",They capture the cases in which the
predicate is true for all possible values in the universe,and in which the predicate is true for at least one value in
the universe,
Slide 3.5.6
Now we are ready to put all these tools together to prove things
about predicates,We are particularly interested in a proof
method called proof by induction,which is shown here,
Less formally,this method basically says the following,
Suppose you can show that the predicate is true for some initial
case (here we are assuming the universe is the set of non
negative integers,so this initial case is n equal 0,if the universe
were different the initial case would also be different),Now
suppose you can somehow also show that for any value of n in
the universe,the implication that if the proposition is true for
that n it is then also true for an incremented n is true,In this
case,induction allows you to deduce that the proposition is true
for all values of n in the universe,
Note what this buys you,It says that you need only show that the statement is true for some base case; and you
need only show that for some arbitrary value of n,if the proposition is true for that n then you can show it is true
for the next value of n,You don't need to check all values; you just need to show that the implication holds,In
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
essence,you can "bootstrap" a proof that the statement is always true,
Slide 3.5.7
Let's make this a bit less murky with an example,Suppose I
want to prove the predicate shown here for all non-negative
values of n,that is,that the sum of the powers of 2 can be
captured by the simple expression shown on the right,
To prove this by induction,I should first really specify the
universe,which I will take to be all non-negative integers,
Then,I need to show that this is true for the base case,when n
is zero,That is easy,I just substitute for n on both sides of the
equation,and check that 1 is the same as 2 - 1,
Now for the inductive step,Let's start with the left side of the
equation for the n+1 case,I can rewrite this as shown in the first
step,Now,I want to rewrite this way because the first term on
the right is just the same predicate,but now for argument n,But by induction,I simply want to show that if this
version of the predicate is true,so is the version I am considering,So I can replace that summation with the
implication from the predicate,and then some simple arithmetic shows that I get the expression I want,I have now
shown that if the proposition is true for n,then it must also be true for n+1,And now by induction I can conclude it
is true for all legal values of n,
Slide 3.5.8
So what we have is a way of proving things by induction,The
steps are as shown,
We first define the predicate we want to prove correct,
including specifying what the variable denotes and the range of
legal values of the variable,
We then prove that the predicate is true in the base case (which
we have here assumed is for n equal to 0)
We then prove that the implication holds,and this allows us to
conclude that the predicate is always true,
Slide 3.5.9
So finally we can relate this back to programs,Suppose we
want to prove that our code for factorial is correct,that is,that it
will always run correctly,Note that there is a hidden
assumption here,namely that we give it a correct input,Note
that here,we have made the unstated assumption that factorial
only applies to integers greater than or equal to one,If we
provide some other value,all bets are off,In other words,our
universe here is positive integers,
So here is our code,and our proposition is that this code
correctly computes n! given a legal value for n,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.5.10
Thus to prove it we first consider the base case,Remember that
here our base case is n equal 1,since that is the starting point
for our universe of values,Is our proposition valid?
Sure! We know that our "if" statement will in this case take the
path through the consequent,and just return the value 1,which
we also know is the desired answer,
Slide 3.5.11
So what about the inductive step? In this case we can assume
that our code works correctly for some arbitrary,but legal,
value of n,and we want to show that under that assumption,it
also works correctly for n+1,
In this case,we will take the consequent path through the "if"
expression,Our substitution model says that the value of this
expression is found by finding the values of the subexpressions,
then applying the first to the others,Finding the value of * is
easy,as is the value of the sum,By induction,we can assume
that the value of the last subexpression will be the correct value
of factorial,
As a consequence,applying the multiplication operator to the
value of n+1 and n! will return (n+1)! and hence we can conclude that this will always produce the right answer!
Slide 3.5.12
Now,let's pull this together,The message to take away from
this exercise is that induction provides a basis for
understanding,analyzing and proving correctness of recursive
procedure definitions,
Moreover,exactly this kind of thinking can be used when we
design programs,not just when we analyze them,When given a
new problem to solve,it is valuable to identify the base case
and find a solution for it,Then,one turns to the issue of
breaking the problem down into a simpler version of the same
problem,assuming that the code will solve that version,and
using that to construct the inductive step,how to solve the full
version of the problem,
We hope that you will use this approach as you build your own procedures,
6.001 Notes,Section 3.1
Slide 3.1.1
In this lecture,we are going to put together the basic pieces of
Scheme that we introduced in the previous lecture,in order to
start capturing computational processes inside procedures,To
do this,we will introduce a model of computation,called the
substitution model,We will show how that model helps us
relate the choices we make in designing a procedure to the
actual process of evaluation that will occur when we use that
procedure,
Thus we will first look at an example of the substitution model,
to make sure you understand its mechanics,and then we will
use to that examine two different approaches to creating
procedures,
Slide 3.1.2
Now that we have the ability to create procedures and use them,
we need some way of figuring out what happens during the
evolution of a procedure application,For that we have
something called the substitution model,You've actually seen
this,but we just didn't call it that,and now we are going to
make it quite explicit,
The role of the substitution model is to provide us with a means
of determining how an expression evolves,In examining the
substitution model,we stress that this is not a perfect model of
what goes on inside the computer,In fact,later in the term we
will see a much more detailed and accurate model of
computation,but this model suffices for our purposes,Indeed,
we don't really care about using the substitution model to figure out the actual value of an expression,We can use
the computer to do that,Rather,we want to use the substitution model to understand the process of evaluation,so
that we can use that understanding to reason about choices in designing procedures,Given that understanding,we
can work backwards,using a desired pattern of evaluation to help us design the correct procedure to generate that
evaluation pattern,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.1.3
So here are the rules for the substitution model,Given an
expression,we want to determine its value,If the expression is
a simple expression there are several possibilities,If it is a self-
evaluating expression,like a number,we just return that value,
If the expression is a name (something we created with a define
expression) then we replace the name with the corresponding
value associated with it,
If the expression is a special form,there are also several
possibilities,If the expression is a lambda then we replace the
expression with some representation of the actual procedure,If
the expression is some other special form (such as an if
expression) then we follow specific rules for evaluating the
subexpressions of that special form,
Otherwise the expression is a compound expression (one of those expressions nested in parentheses that contains
several subexpressions),Here we apply exactly the same set of rules to each of the subexpressions of the
compound expression,in some arbitrary order,We then look at the value of the first subexpression,If it is a
primitive procedure (one of the built-in procedures,like * or +) then we just apply that procedure to the values of
the other expressions,On the other hand,if the value of the first subexpression is a compound procedure
(something we created by evaluating a lambda expression) then we substitute the value of each subsequent
subexpression for the corresponding procedure parameter in the body of the procedure,We replace the entire
expression with this instantiated body,and repeat the process,
Slide 3.1.4
So here is a simple example,Suppose we have defined square
to be the obvious procedure that multiplies a value by itself,
Now,let's use the substitution model to trace out the evolution
of applying that procedure,
The substitution model says that to trace the evaluation of
(square 4),we first determine that this is a compound
expression,Since it is,we evaluate each of the subexpressions
using the same rules,Thus,the first subexpression,square,is
just a name,and we look up its value,getting back the
procedure we created when we evaluated the lambda,The
second subexpression is just a number,so that is its value,Now
the rule says to substitute 4 everywhere we see the formal
parameter x in the body of the procedure and replace the original expression with this instantiated body expression,
as shown,
This reduces the evaluation of (square 4) to the evaluation of the expression (* 4 4),This is also a compound
expression,so again we get the values of the subexpressions,In this case,the application is of a primitive
procedure,so we simply apply it to reduce the expression to its final value,16,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.1.5
Here is a more detailed example,First,let's create two
procedures,These expressions will each generate a procedure,
through the lambda expressions,and the defines will then
associate a name with each of them,Thus,in our environment,
we have pairings of square and average to the associated
procedure objects,each with its own parameter list and body,
Slide 3.1.6
Here is an example using these procedures,To evaluate this
first expression using the substitution model,I first need to get
the values of the sub-expressions,The value of average I get by
looking this name up in the environment,giving me the
associated procedure object,The value of 5 is easy,The last sub-
expression is itself a combination,so I recursively apply the
rules,By the same reasoning I get the procedure associated with
square and the value 3,then substitute 3 into the body of square
for x,This reduces to another combination,which I can
recognize is an application of a primitive procedure,so this
reduces to the number 9,Thus the recursive application of the
substitution model yields a simpler combination,
Note the format here,I first evaluate the sub-expressions,then substitute and apply the procedure,This is called an
applicative order evaluation model,
Slide 3.1.7
Notice in particular that under this model I need to get the value
of the operands first,which caused me to evaluate that interior
compound expression before I got around to applying average,
Once I have simple values,I can continue the substitution,In
this case,I use the body of the procedure associated with
average,now substituting 5 and 9 for x and y in that body,
Note that there is no confusion about which x I am referring to,
it's the one in the procedure associated with average,not the
one in square,
As before,I substitute into the body of this procedure,and then
continue,I recursively apply the rules to each subexpression,In
the case of a compound expression,I must reduce this to a
simpler value before I can continue with the rest of the expression,Also note that when the application involves a
primitive procedure,I just do the associated operation,eventually reducing this to the final answer,
The key idea is to see how this notion of substitution lets us trace out the evolution of an expression's evaluation,
reducing an application of a procedure to the evaluation of a simpler expression,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
6.001 Notes,Section 3.2
Slide 3.2.1
Now let's look at a little less trivial example,Suppose I want to
compute factorial of n,which is defined as the product of n by
n-1 and so on down to 1,Note that I am assuming that n is an
integer and is greater than or equal to 1,
Slide 3.2.2
So how do I create a procedure to compute factorial of n? Well
I need to think carefully here,since my choice in designing this
procedure will impact the evolution of the process when I use
it,
One way to look at factorial is to group the computation into
multiplying n by all the other multiplications,but I can
recognize that this latter computation is just factorial of n-1,
So notice what this does,it reduces a computation to a simpler
operation,multiplication,and a simpler version of the same
problem,in this case factorial of a smaller number,So I have
reduced one version of a problem to a simpler version of the
same problem,plus some simple operations,This is a very
common pattern that we are going to use a lot in designing procedures,
Slide 3.2.3
In fact,given this common pattern,we can write a procedure to
capture this idea,Here it is,The first part says to give the name
fact to the procedure created by the lambda,which has a single
argument n and a particular body,Now,what does the lambda
say to do? It has two different parts,which we need to look at
carefully,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.2.4
First of all,it has a new kind of expression in the interior,the =,
This is an example of a predicate,a procedure that takes some
arguments and returns either true or false,in this case based on
the test for numerical equality,
Slide 3.2.5
The more interesting part is the body of the lambda,and it is a
new special form called an if,Because this is a special form,it
means that evaluation does not follow the normal rules for
combinations,If's are used to control the order of evaluation,
and contain three pieces,
Slide 3.2.6
Its first subexpression we call a predicate,
Slide 3.2.7
Its second subexpression we call a consequent,.,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.2.8
..,and its third subexpression we call an alternative,Now here
is why an if is a special form,An if expression first evaluates its
predicate,using the same set of rules recursively on this
expression,It does this before it ever looks at either of the two
other expressions,Now,if the predicate evaluates to a true
value,then the if expression takes the consequent,and evaluates
it,returning that value as the value of the overall if expression,
On the other hand,if the predicate evaluates to a false value,
then the if expression takes the alternative expression,evaluates
it,and returns that value,Notice that only one of the consequent
and alternative expressions is evaluated during an if,
In the case of our fact procedure,we can now see how the
evaluation will proceed,When we apply fact to some argument,we first test to see if that value is 1,The if
expression does this by first evaluating the predicate,before ever considering the other expressions,thus changing
the order of evaluation,If the predicate is true,then we simply return the value 1,If it is false,then we will
evaluate the last expression with appropriate substitution,thus unwinding factorial of n into the multiplication of
the value of n by the result of evaluating factorial of n-1,
Slide 3.2.9
Here is a tracing of the substitution model in action,To
evaluate factorial of 3,we substitute into the body of fact,
leading to the next expression,The if first evaluates the
predicate,leading to the next expression,Since the predicate's
value is false,the entire if expression is replaced by the
alternative,with appropriate substitutions,
To evaluate this expression we need to get the values of the
subexpressions,so this entire expression reduces to a
multiplication of 3 by the value of (fact 2),Notice how this has
unwound the computation into a particular form,a deferred
multiplication and a recursive call to a simpler version of the
same problem,
This process repeats,using the if to unwind another level of the computation,and this continues,just using the
substitution model rules,until finally the predicate is true,In this case,we are left with an expression just involving
primitives,and we can complete the multiplications,thus reducing the computation to a simple answer,
Note the form shown in red,in which an expression is reduced to this pattern of a simpler operation and a simpler
version of the same problem,This unwrapping continues until we are left with a nested expression whose
innermost subexpression only involves simple expressions and operations,at which point the deferred operations
are evaluated,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.2.10
That means that fact gives rise to what we call a recursive
process,In the substitution model we can see that this is
characterized by a set of deferred operations,in which the
multiplication is deferred while we go off to get the sub-
computation of the simpler version of fact,Once we get to a
simple case,we can start accumulating those stacked up
operations,We will come back to this idea next time,
6.001 Notes,Section 3.3
Slide 3.3.1
Now that we have seen our first more interesting procedure,
fact,let's step back and generalize the ideas we used to capture
this computation in a procedure,
What are the general steps that we use in creating a recursive
procedure like this? We have three stages,as shown here,
The first stage is called wishful thinking,The idea is as
follows,If I want to create a solution to some problem,I first
assume that I have available a procedure that will solve the
problem,but only for versions of the problem smaller than the
current one,In the case of factorial,I assume (wishfully) that I
can solve factorial for problems of size smaller than n,
Slide 3.3.2
Given that assumption,the second stage proceeds to design a
solution to the problem,In particular,I can use the existence of
a solution to the smaller sized problem,plus a set of simple
operations,to design the solution to the larger sized problem,
We saw this with factorial,where we combined the solution to
a smaller version of factorial,plus a multiplication,to create the
solution to the full version of factorial,
Notice that the second step here requires some ingenuity,One
should be careful to think through this strategy,before
beginning to code up a solution,In the case of factorial,we saw
how we did this decomposition into a multiplication and a
smaller version of factorial,
With this,we can build a first version of the procedure as shown at the bottom of the slide,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.3.3
Of course you already know that this won't quite work,but let's
think about why,If we apply that version of fact to some
argument,it will keep unwinding the application of factorial
into a multiplication and a smaller version of factorial,But
before we can compute that multiplication,we have to get the
value of the smaller version of factorial,and that will continue
to unwind to another version,ad infinitum,
The problem is that I didn't use my third step; I didn't consider
the smallest sized problem that I can solve directly,without
using wishful thinking,In the case of factorial,that is just
knowing that factorial of 1 is just 1,Once I find that smallest
sized problem,I can control the unwinding of the computation,
by checking for that case and doing the direct computation when appropriate,In this way,I will terminate the
recursive unwinding of the computation and accumulate all the deferred operations,
Slide 3.3.4
This in fact is a very common form,a common pattern for a
recursive algorithm that has a test,a base case and a recursive
case,The if controls the order of evaluation to decide whether
we have the base case or the recursive case,The recursive case
controls the unwinding of the computation,doing so until we
reach the base case,
Slide 3.3.5
To summarize,we have now seen how to design recursive
algorithms,by using this idea of wishful thinking to separate
the problem into a simpler version of the same problem,do that
decomposition and identify the smallest version of the problem
that will stop the unwinding of the computation into simpler
versions of the same problem,As a consequence,algorithms
associated with this kind of approach have a test,a base case,
and a recursive case to control the order of evaluation in order
to unwind the computation into simpler versions of the same
problem,
6.001 Notes,Section 3.4
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.1
So we have introduced recursive procedures,and we have now
seen an example of building a recursive procedure to implement
factorial,Now,let's look at a different kind of procedure,an
iterative procedure for computing factorial,In doing this,we
are going to see how we can develop procedures with different
evolutions that compute the same basic computation,
Slide 3.4.2
To set the stage for this,here is our factorial procedure from
last time,Remember that this was a nice little procedure that
first checked to see if we were in the base case,and if we were,
returned the answer 1,If we were not,the procedure reduced
the problem to multiplying n by a recursive call to factorial of
n-1,One of the properties of this procedure was that it held a
set of deferred operations,To compute factorial of n,it had to
hold onto a multiplication while it went off to compute the
subproblem of a simpler version of factorial,and that in turn
required another deferred operation,Thus,the procedure had to
keep track of something to do,once it was done computing a
subproblem,and this meant that as the argument to the procedure gets larger,we would have more things to keep
track of,We would like to see if there is another way of computing factorial,without all of those deferred
operations,
Slide 3.4.3
The specific question we want to consider is whether we can
design an algorithm for factorial that uses only constant space,
By this we mean that the computation does not require keeping
track of any deferred operations,but rather can perform its
work without using up any additional memory to keep track of
things to do,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.4
So that's our goal,to compute factorial in a different way that is
not characterized by those deferred multiplications
Slide 3.4.5
..,and here is the intuition behind this different approach,Let's
think about how you might do this if you were executing the
computation by hand,One way to do this is to multiply out the
first two numbers,and keep track of that temporary product,
plus the next stage in the computation,For example,for
factorial of 4,you would multiply out 4 by 3,keep track of the
result,12,and the fact that the next thing to multiply is 2,You
would then multiply the current result,12,by 2,keeping track
of that result and the next thing to multiply by,
Slide 3.4.6
Notice that in this version,we only need to keep track of 2
things,the current product,and the next thing to do,This is
different from the recursive case,where we were blindly
keeping track of each pending multiplication,Here,we replace
the previous product by the new one,and the next multiplier by
the new one,whereas in the previous case we had to keep track
of each deferred operation,i.e,I have to multiply by 2,then I
have to multiply by 3,then,.,
Slide 3.4.7
Thus,we only need a constant amount of space,To stress,think
about the recursive case,There,we had to write down that I
have a pending operation of multiplication by some number,
plus another pending operation of multiplication by another
number,and so on,Keeping track of each such pending
operation takes a bit more space,In this case,I only need two
pieces of information,the current product and the next
multiplier,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.8
The reason I can do this is because multiplication satisfies two
nice properties from mathematics,that in essence say I can do
the multiplication in any order,and accumulating partial
answers along the way will not affect the result,Thus,I can
come up with an algorithm that only uses constant space,
Slide 3.4.9
So now let's capture that idea a little more formally,I can use
the same sort of notion in a table,in which I have one column
for each piece of information I am going to need,and one row
for each step of the computation,Each step here means how to
do I go from a current state,captured by a temporary product
and a counter that tells me the next thing to multiply,to the next
state of those parameters,
Slide 3.4.10
Now what I need to do is come up with a rule for how I am
going to change the values in the table,In particular,to get the
next value for product,I take the current value of product and
the current value of counter,multiply them,and that becomes
the next entry in the product column,
Slide 3.4.11
Similarly,I need a rule for getting the next value for counter,
That,we know,is simply given by adding 1 to the current row's
value for counter,thus generating the next row's value for
counter,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.12
Of course,I can keep repeating this process,updating the
values for the product column and the counter column until I
am ready to stop,
Slide 3.4.13
And how do I know when I am ready to stop? Well that's easy,
The last row I want in the table is the one when counter is
bigger than n,
Slide 3.4.14
Ah -- and that indicates one more thing,I will need another
column to keep track of n,since each column specifies a
needed piece of information and that is one more,The fact that
the value doesn't change is irrelevant,what matters is that I
need to keep track of this information,
Slide 3.4.15
Once I am done,I still need to get the answer out,but I know
where that is,It's sitting in the element at the intersection of the
product column and the last row,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.16
The only other thing I need to do is figure out how to start this
process off,And this is also easy,My first row will deal with
the equivalent of a base case,0! is just 1,so I can start my
product at 1,my counter at 1,set up N to be whatever value I
want to compute,and I can then start this table process
underway using the rules I just described,
Slide 3.4.17
Now let's write a procedure to do this,I am going to define a
procedure called ifact,As you can see,I use a lambda
to create the procedure of one argument,and I use a define
to give it a name,Notice that the body of the procedure is just a
call to another procedure,a helper procedure that I am also
going to define,
The reason I need a helper procedure is clear from my previous
discussion,I need to keep track of three things during the
computation,so I need a procedure with three arguments,as
shown,one for each column of my table,The body of this
helper procedure looks a lot like my earlier recursive procedure,
but with some important differences,First,there is a test case,to tell when we are done,Here we don't have a base
case,but rather there is a return value,an answer to give when the test case holds,The iterative case,the thing to
do if the test case is not true,is slightly different from our recursive case,Here,the call to the procedure has new
arguments,one for each parameter,Notice how the new argument in each case captures the update rule from the
table,
Slide 3.4.18
So now let's think about what happens if I evaluate ifact,
The value of ifact with some argument n will be the same
as evaluating the body of the procedure,that is ifact
helper with arguments 1,1,and n,Notice that this is just
setting up the initial row of my table,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.19
Then look at what happens when we evaluate ifact
helper,In fact,it just computes the next row of that table,
It says that the value of ifact-helper when called on
arguments for product,counter and n,is the same as evaluating
ifact-helper with new values for product,counter and
n,each of which exactly captures the update rules from my
table,
Slide 3.4.20
And what else do I need? Just to determine when to stop,So I
have a test to determine when that happens,and at that stage,I
know that the answer is in the product column,so I simply
return that value,
ifact on some argument,say 4,reduces to calling ifact-
Slide 3.4.21
To trace how this works,we use our substitution model,Calling
helper on arguments 1,1,and 4,since that is what the
body of ifact contains,where we have substituted 4 for
the parameter n,
Slide 3.4.22
The substitution model then says to substitute the arguments for
the formal parameters in the body of ifact-helper and
evaluate that instantiated body,namely this if expression,We
saw that if is a special form,and this means we first evaluate
the predicate,Since in this case it returns a false value,the
value of the whole if expression reduces to the value of the
alternative clause,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.23
Evaluating the expressions for the arguments then reduces to
this expression,
Notice an important thing,We have reduced the evaluation of
one expression involving ifact-helper to a different
expression involving ifact-helper,with a different set
of arguments,or parameters,But there are no deferred
operations here,that is,there are no operations that we have to
keep track of while we go off and compute some subproblem,
Rather the answer to evaluating this expression is the answer to
evaluating the first one directly,
Slide 3.4.24
And we can just repeat this process,with the same behavior,
until the predicate is true,
Notice the form of the red expressions,They are constant in
size,There are no deferred operations,no storage of
expressions while I go get the value of a subexpression,Instead,
the variables that capture the state of the computation change
with each evaluation,This is very different from the shape of
the previous version of factorial,There,we had a set of deferred
operations,and the size of the expressions grew with each step,
Here there are no deferred operations,and a constant amount of
space is needed to keep track of the computation,
Slide 3.4.25
Let's compare this to the recursive version,Remember that we
had that pending operation,and it caused the shape of the
procedure's evolution to be different,It grew with each new
evaluation,because that extra pending operation had to be
tracked,including information relevant to each operation,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.4.26
Compare that to the iterative version,Here there are no pending
operations,The value of one call to the procedure is the same as
the value of a different call to the same procedure,that is,
different arguments but nothing deferred,As a consequence,
this procedure has a fixed size,or constant,evolution,
So we have seen two different implementations of the same
computation,with different behaviors,
Slide 3.4.27
And this summarizes what we have seen so far,
6.001 Notes,Section 3.5
Slide 3.5.1
So how do we know that our code is actually correct? It would
be nice if we could guarantee that our code would run correctly,
provided we give it correct inputs,So how do we ensure this?
Well there are several possibilities,We could accept the word
of someone with power over our future (e.g,a 6.001 professor)
that the code is right,Clearly not a wise idea,though it is
shocking how often this approach is used! A more common
method is to rely on statistics,that is,to try our code on a bunch
of example inputs,and hope that if it works on all the ones we
try,it will work on every input,Sometimes this is the best you
can do,but clearly there are no guarantees with this approach,
All too often,programmers rely on the third method,also
known as arrogance,They just assume that they wrote correct code,and are stunned when it fails,
The best method,when possible,is to actually formally prove that our code is correct,And it is this method that we
are going to briefly explore here,We won't show you all the elements of proving programs correct,but we will
provide you with a basis for understanding how this is done,and more importantly for using the reasoning pattern
underlying this proof method to help you design good code,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.5.2
First,we need to talk about what constitutes a formal proof,
Technically,a proof of a mathematical or logical proposition is
a chain or sequence of logical deductions that starts with a base
set of axioms and leads to the proposition we are attempting to
prove,Now,we need to fill in the details on these pieces,
First,a proposition is basically a statement that is either true or
false,Typically,in propositional logic,we have a set of atomic
propositions,i.e.,simple statements of fact,For example,"n=0"
is an atomic proposition,Given some value for the variable n,
this statement is either true or false (assuming that n is a
number,of course),
More interesting propositions (and things that we are more
likely to be interested in proving) are created by combining simpler propositions,There are five standard ways of
creating compound propositions,one can take the conjunction (which acts like an "and",meaning the compound
proposition is true if and only if both of the individual propositions are true); one can take the disjunction (which
acts like an "or",meaning the compound proposition is true if and only if one or the other of the individual
propositions is true); one can take the negation of a proposition (which means the proposition is true if and only if
the initial proposition is false); we can create the implication (which means that if P is true,then Q is true); and we
can establish an equivalence (which means that P is true if and only if Q is true),
Note that compound propositions can have elements that are themselves compound propositions,so that we can
create deeply nested propositions,
Slide 3.5.3
We can capture the logical structure of these compound
propositions by simply looking at the possible cases,If we
consider all possible combinations for the base propositions,we
can chart out the veracity of the compound proposition,Thus,
we see that "P and Q" behaves as we would expect; it is true as
a statement if and only if both P and Q are true,Negation and
disjunction also behave as expected,So does equivalence,since
the combination can only be true when both simpler elements
are,
Implication is a bit more puzzling,The proposition is defined to
be true if either P is false or Q is true,To see this,take a simple
example,Suppose we consider the proposition,"If n is greater
than 2,then n squared is greater than 4",Then for example,if n equals 4,both propositions are true,and our truth
table also states that the implication is true,If for example,n equals -4,then the first proposition is false and the
second proposition is true,The compound statement is true,as we can see from the truth table,and this makes
sense,since the consequent can clearly still be true even if the precedent is not,Finally,suppose n equals 1,In this
case both propositions are false,however the statement as a whole is true,
This particular combination thus has a rather counterintuitive aspect,it is always true if the first part is false,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.5.4
Now,given that we can construct propositions,we want to be
able to prove their correctness by chaining together a sequence
of logical deductions,starting from a set of axioms (which are
propositions that are assumed to be true),The basic idea is that
we start with statements that are correct,and then use standard
deductions to reason about statements whose veracity we are
attempting to establish,
For example,a common inference rule (which has been around
at least since Aristotle) is modus ponens,This basically
says that if we have a proposition P that is true,And we have an
implication (P implies Q) that is also true,and then we can
deduce that Q is a true proposition,
A related (and equally old) inference rule is modus tollens,This basically states that if we have an
implication that P implies Q,and we also know that Q is not true,then we can infer that P is not true,
Slide 3.5.5
Now,given these tools,we could go off and prove statements
of fact,Unfortunately,the things we want to prove about
programs (e.g,does this code run for all correct inputs) require
potentially infinite propositions,since we would have to take a
conjunction of propositions,one for each input value,So we
need to extend our propositional logic to predicate logic,
We handle this by creating propositions with variables,and
then specifying the set of values (called the universe or domain
of discourse) over which the variable may range,
Since predicates apply over ranges,we also can put conditions
on those predicates,The two basic quantifiers are known as
"for all" and "there exists",They capture the cases in which the
predicate is true for all possible values in the universe,and in which the predicate is true for at least one value in
the universe,
Slide 3.5.6
Now we are ready to put all these tools together to prove things
about predicates,We are particularly interested in a proof
method called proof by induction,which is shown here,
Less formally,this method basically says the following,
Suppose you can show that the predicate is true for some initial
case (here we are assuming the universe is the set of non
negative integers,so this initial case is n equal 0,if the universe
were different the initial case would also be different),Now
suppose you can somehow also show that for any value of n in
the universe,the implication that if the proposition is true for
that n it is then also true for an incremented n is true,In this
case,induction allows you to deduce that the proposition is true
for all values of n in the universe,
Note what this buys you,It says that you need only show that the statement is true for some base case; and you
need only show that for some arbitrary value of n,if the proposition is true for that n then you can show it is true
for the next value of n,You don't need to check all values; you just need to show that the implication holds,In
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
essence,you can "bootstrap" a proof that the statement is always true,
Slide 3.5.7
Let's make this a bit less murky with an example,Suppose I
want to prove the predicate shown here for all non-negative
values of n,that is,that the sum of the powers of 2 can be
captured by the simple expression shown on the right,
To prove this by induction,I should first really specify the
universe,which I will take to be all non-negative integers,
Then,I need to show that this is true for the base case,when n
is zero,That is easy,I just substitute for n on both sides of the
equation,and check that 1 is the same as 2 - 1,
Now for the inductive step,Let's start with the left side of the
equation for the n+1 case,I can rewrite this as shown in the first
step,Now,I want to rewrite this way because the first term on
the right is just the same predicate,but now for argument n,But by induction,I simply want to show that if this
version of the predicate is true,so is the version I am considering,So I can replace that summation with the
implication from the predicate,and then some simple arithmetic shows that I get the expression I want,I have now
shown that if the proposition is true for n,then it must also be true for n+1,And now by induction I can conclude it
is true for all legal values of n,
Slide 3.5.8
So what we have is a way of proving things by induction,The
steps are as shown,
We first define the predicate we want to prove correct,
including specifying what the variable denotes and the range of
legal values of the variable,
We then prove that the predicate is true in the base case (which
we have here assumed is for n equal to 0)
We then prove that the implication holds,and this allows us to
conclude that the predicate is always true,
Slide 3.5.9
So finally we can relate this back to programs,Suppose we
want to prove that our code for factorial is correct,that is,that it
will always run correctly,Note that there is a hidden
assumption here,namely that we give it a correct input,Note
that here,we have made the unstated assumption that factorial
only applies to integers greater than or equal to one,If we
provide some other value,all bets are off,In other words,our
universe here is positive integers,
So here is our code,and our proposition is that this code
correctly computes n! given a legal value for n,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 3.5.10
Thus to prove it we first consider the base case,Remember that
here our base case is n equal 1,since that is the starting point
for our universe of values,Is our proposition valid?
Sure! We know that our "if" statement will in this case take the
path through the consequent,and just return the value 1,which
we also know is the desired answer,
Slide 3.5.11
So what about the inductive step? In this case we can assume
that our code works correctly for some arbitrary,but legal,
value of n,and we want to show that under that assumption,it
also works correctly for n+1,
In this case,we will take the consequent path through the "if"
expression,Our substitution model says that the value of this
expression is found by finding the values of the subexpressions,
then applying the first to the others,Finding the value of * is
easy,as is the value of the sum,By induction,we can assume
that the value of the last subexpression will be the correct value
of factorial,
As a consequence,applying the multiplication operator to the
value of n+1 and n! will return (n+1)! and hence we can conclude that this will always produce the right answer!
Slide 3.5.12
Now,let's pull this together,The message to take away from
this exercise is that induction provides a basis for
understanding,analyzing and proving correctness of recursive
procedure definitions,
Moreover,exactly this kind of thinking can be used when we
design programs,not just when we analyze them,When given a
new problem to solve,it is valuable to identify the base case
and find a solution for it,Then,one turns to the issue of
breaking the problem down into a simpler version of the same
problem,assuming that the code will solve that version,and
using that to construct the inductive step,how to solve the full
version of the problem,
We hope that you will use this approach as you build your own procedures,
