[Contents]
Copyright © 2014 jsd

Problem-Solving Checklist
John Denker

*   Contents

1  Prerequisites, Prior to Problem-Solving

Consider the following parable:

Once upon a time there were two farmers, Simplicio and Salvatio. They were very similarly situated. They operated neighboring tomato farms. Their fields were nearly the same size.

    
On harvest day, Simplicio said “I’m going to work really hard today. I’m going to work so hard that I harvest twice as many tomatoes as the other guy. I planted twice as many plants, so there should be no problem. I’m going to turn a huge profit.”

    
However, there is one slight problem. Because the plants were too close together, they weakened each other. Then the worms came along and finished them off. As a result, there are hardly any tomatoes in Simplicio’s field. It doesn’t matter how hard he works on harvest day; he can’t harvest tomatoes that aren’t there.

Problem-solving day is like harvest day. You get to harvest and profit from facts and ideas that are in your brain. The process is subject to a crucial constraint: You cannot harvest stuff that isn’t there. If you didn’t put ideas into your head in the proper way, you won’t be able to get them out when needed, no matter how hard you work at it.

Let’s be clear: In large measure, your success at problem-solving depends on what you did during previous days, months, and years. You have to learn stuff over time. There are right ways and wrong ways to do this; see e.g. item 1.

When reading this document – or any other writeup on problem-solving – please be aware of its limitations. There is nothing wrong with what is said here; the problem is what is not said. There is some value in problem-solving techniques per se, but you must look elsewhere to find the other 95% of what you need for effective learning, thinking, and problem-solving.

Here’s another limitation: A checklist such as this one (or reference 1) can be useful as a reminder of things you already mostly know. However, it’s hard to learn new ideas from a checklist. It’s too condensed. In some cases, an individual item on this list (e.g. item 46) would become a book chapter or an entire book if spelled out. If you want to learn, it helps to have more detail. It helps to see the ideas applied to examples. There are quite a few good examples in reference 2. An additional nice example is discussed in reference 3.

2  General Suggestions

1.
Whenever you encounter a new idea, mull it over. Think about all the ways in which it is connected to – or inconsistent with – things you already know.

This applies all day every day, whether you are trying to solve a particular problem or not. It applies to every new idea you encounter, whether it comes from something you read, or an experiment you did, or a calculation you did, or whatever. This is discussed in some detail in reference 4.

The connections are necessary if the idea is to be useful, because they are what allow you to recall the idea when needed. To say the same thing the other way, if you put an idea into your brain but have no way of recalling it when needed, it might as well not be there. An idea that can be recalled in many ways is vastly more useful than one that can only be recalled in one way.

2.
As a related point: Even when you’re not working on a specific problem, look for patterns in the world around you. This might lead you to formulate a good question or a hypothesis that needs testing. Conversely, always be on the lookout for anomalies, for deviations from the expected pattern.

This is important for two reasons:

3.
Don’t panic.

4.
Check the work.

In more detail:

Following this advice may slightly increase the workload in the short run, but it greatly reduces the workload in the long run, especially when dealing with complex problems. For details on this, see reference 5.

Checking the work at each step along the way greatly increases the odds of getting the right answer. In a complex calculation, there will be mistakes, and showing the work makes it easier to find and fix the mistakes. Showing the work and showing the checks makes it easier to check and re-check the work. It is particularly valuable in a teamwork situation, because it helps other people check the work and learn from the work. For more on this, see section 4 and item 15.

Check The Work.
     

5.
Don’t believe everything you read. This can be considered a corollary of item 1. It can also be considered a corollary of item 4, insofar as you are checking the other guy’s work.

6.
Document the code. This is a corollary of item 4. It shows how the code was designed, which greatly facilitates checking. This is discussed in reference 6.

7.
Write out a glossary, spelling out what your notation means, and what your terminology means. Do not rely on the name of an idea encode the full meaning the idea. We routinely give short names to complex things; that’s what names are for. For more on this, see reference 6.

This is particularly important when you are working as part of a problem-solving team. You are free to make up whatever notation and terminology you like, but don’t expect other people to guess what it means.

8.
Draw the diagram.

There exist thousands of different types of diagrams; reference 7 lists more than 100 examples. If you cannot find a pre-existing way of representing the data, invent something. If this is not easy, do it anyway. Remember the sardonic proverb:

When all else fails,
look at the data.
     

9.
When given an assignment, make sure you understand the statement of the problem, so you don’t go off and solve the wrong problem. See also item 10.

10.
When given an assignment, check to see how ill-posed it is.

Procedures for dealing with ill-posed problems are discussed in reference 8.

11.
Formalism can be very powerful, and sometimes it delivers a valid result whether you understand it or not.

You are absolutely not required to fully understand every element of what’s going on before starting the analysis and calculations. It is common to understand a problem well enough to solve it, and then achieve a much deeper understanding by pondering the solution. See also the discussion of formalism in section 5.

12.
Sometimes experiment delivers a valid result, whether you understand it or not. It is fairly common to obtain a experimental result and not fully understand it until later. There are layers upon layers of understanding. There is such a thing as serendipity.

13.
During the process of solving the problem, build up an understanding of what’s going on (as far as possible). That is to say, don’t blindly follow the formalism (any more than necessary).

Analysis helps with the understanding ... just as understanding helps with the analysis. You can leverage one against the other, again and again, recursively and iteratively, like the itsy-bitsy spider.

Even though it is sometimes possible (as mentioned in item 11 and item 12) to obtain a formally-correct result without understanding it, this should be considered only a first step. It is not something to be proud of. Don’t walk away from the result until you have made a serious effort to understand it. See also item 14 and item 17.

14.
If a complicated derivation produces a simple result, go back and look for a simpler derivation. This is part of the process of understanding, as mentioned in item 13.

15.
After you have found one solution, see if you can find another solution to the same problem. Try to make the solutions independent, to the extent possible. If they have some steps in common, that’s not tragic, but try to minimize the overlap.

Example: There are two remarkably dissimilar solutions to the Mississippi Flow problem (reference 4).

Example: There are two remarkably dissimilar solutions for finding the Johnson noise in a parallel combination of resistors (reference 9).

Obviously this overlaps quite a bit with the suggestion to Check The Work (item 4) but it is worth mentioning, because it is a specific way of checking the work. The obvious approach is to carry out the same calculation a second time, checking and re-checking each step along the way.

The non-obvious approach is to carry out a completely different calculation that produces the same result. This has the tremendous advantage that it can catch conceptual errors, not just arithmetic errors. (The downside is that if there is an error, this will not tell you exactly where to find it.)

Simple example: Check by doing the inverse problem. Given the task 10 − 7 = ____, the answer plus 7 had better equal 10.

Intermediate example: If you use trig functions to find the sides of a triangle, the answer had better uphold the Pythagorean theorem and other trig identities. Similarly, any physics answer had better uphold the various conservation laws.

Fancy example: Given the task of estimating (closed book) how much Mississippi River water flows past New Orleans in a year, there are at least two completely different ways of approaching the problem, leading to completely independent estimates.

For students who have never seen the problem before, finding one solution is hard ... and finding two independent solutions is even harder, much worse than twice as hard.

16.
If you don’t immediately see how to get started on the assigned problem, it may help to do some warm-up exercises. That is, attack a similar but simpler problem.

17.
After you have solved the assigned problem, see if you can generalize it. This is an important part of building up an understanding, as discussed in item 13.

18.
Given a word problem, find a way to formalize it. This often involves choosing a set of variables, and then writing some equations that express relationships between the variables. This in turn may involve choosing a suitable frame of reference, system of coordinates, vector basis, et cetera.

19.
Don’t hesitate to break one hard problem into a great many easy problems. You don’t need to be Superman. That is to say, you don’t need to be able to leap tall buildings at a single bound if you can walk up the stairs. There are a lot of steps, but each one is easy. See also section 3.

For example, balancing a complicated chemical reaction equation would be well-nigh impossible if you didn’t have a systematic way of doing it. Gaussian elimination might require 20 steps or even 100 steps, but each step is cut-and-dried and incredibly easy, so you just grind it out. Or tell the computer to do it. Problem solved.

20.
Attack the problem from all angles. If necessary, chip away at the corners. Try something and see if it works. If not, try something else. Use your intuition to help decide what is a plausible line of attack. See also section 3.

Feynman compared knowledge to a grand tapestry. Something you don’t know is like a hole in the tapestry. You can fix it by weaving upward from the bottom, downward from the top, and/or inward from the sides. Sometimes it pays to start in the middle of nowhere and construct something from scratch, and then patch it into the main tapestry later.

21.
Beware that a great many people who ought to know better emphasize the straight-backwards attack, i.e. starting with the conclusion and working backwards. This is no worse than many of the other possible lines of attack, but it is no better, and there are thousands of possibilities.

The straight-backward approach works particularly well for some of the multiple-guess tests that plague our classrooms today, but does not work nearly so well in the real world.

22.
As emphasized by Pólya (reference 2), pay special attention to the hardest subcase you can solve, and the easiest subcase that you cannot (yet) solve.

23.
Paper is cheap. Don’t try to cram calculations into a corner of the page; start a new page. Don’t skip steps in an attempt to save paper. See reference 11.

24.
When you use one equation to derive another, you should not do it by revising the initial equation in place. Instead, leave the original equation alone, and write the derived equation on a new line. This is important for making the work checkable. See reference 11.

25.
Generally, you should avoid erasing anything in your working notes. If it’s wrong, draw a tidy diagonal line through it, rewrite it correctly if necessary, and proceed from there. Ditto for seemingly-worthless dead ends. Anybody who is reading your work knows what a mistake is and what a dead end is, and won’t be confused or surprised to find such things in your work. Contrast item 26. For additional details, see reference 11.

Rationale: In the best case, erasing isn’t the trouble. In the worst case, you run the risk of erasing something valuable.

26.
If you are keeping your notes on a computer, use a version control system. This allows you to “tidy up” the current version as much as you like. Meanwhile, previous versions can be retrieved at the touch of a button.

27.
Keep track of the difference between the main goal (solving the maze) and the various sub-goals (reconnoitering some branch of the maze). Often making progress toward the main goal requires giving up on a sub-goal (i.e. backing out of a dead end). See reference 4.

3  Piece by Piece, not Step by Step

The “scientific method” is not nearly as methodical as non-scientists seem to think it is.

Thinking of it as a step-by-step process suggests stepping along a road, moving steadily forward along a one-dimensional path. That’s not how problem-solving actually works. That may be a good way to explain the solution once it is found, but that’s not a good way to find it.

The reality is more like searching a maze (rather than following a clear one-dimensional path). It goes something like this:

Start with an intuitive idea. Draw a qualitative diagram. That provides motivation to write some equations, which can be turned into software code, which produces a vastly more accurate quantitative diagram. Looking at this diagram indicates that the original intuitive idea is not entirely satisfactory. Checking the limiting cases shows that the idea works in certain limits, but not in the general case. This suggests a more elaborate idea. That motivates writing some new equations and some new code. That produces a new diagram. Checking the limiting cases and the symmetries suggests that the “elaborate” model can be made much more streamlined and more intuitive. That motivates yet more equations. Alas the equations show that the streamlined model is not actually correct. So we have to back up and come up with yet another model. A new concept is brought to bear. It is sophisticated but not complicated. This motivates an entirely new approach, including new equations, new code, and new diagrams. The result is elegant and provably correct. It is intuitive in retrospect. There is an algebraic proof and a non-algebraic geometric proof, which can be checked against each other. An analysis of the symmetries, limiting cases, and analogies makes it even more plausible and intuitive.

One lesson to take from this story is that a check-list is not a do-list. You should check each of the items enumerated in section 2, but you don’t necessarily need to do them all. You certainly don’t need to do them in any particular order. Some of them will get done multiple times at various stages of the game.

Very commonly you chip away at one corner of the problem and go as far as you can until you get stuck. Then you set that aside and start chipping away at another corner.

Very commonly a seemingly-promising line of attack turns out to be a dead end, and a big chunk of preliminary work has to be abandoned. Abandoning something is painful and difficult for psychological reasons, but it has to be done.

Consider the following styles:

For most of the last 2300 years, there has been a tendency to pretend these are all the same thing ... when in fact they are not. Not even close.

I like elegance and polish as much as the next guy, but not when it requires pulling rabbits out of hats in such a way that the students cannot understand the process.

There will always be enormous temptation to polish things, for some good reasons and some not-so-good reasons.

  1. One semi-justifiable reason is efficiency: Real problem solving involves a lot of back-tracking out of dead ends. It is more efficient if you can skip the dead ends, with the help of 20/20 hindsight.

    The polished method works fine if you are using the solution for the third or fourth time.   The polished method isn’t at all representative of the process of discovering the solution for the first time.

    If all you want is a solution, the elegantly polished solution is fine.   If you want to learn how to solve problems, you need to see the problem-solving process, which is rarely (if ever) elegant.

    When teaching, or when writing a textbook, you have to decide whether you are training solution-users or educating solution-discoverers. It isn’t practical to show the gory details every time, but I suggest showing them at least occasionally.

  2. Another factor is that people always like to pretend they are smarter than they really are. The temptation is to say “here is my elegant solution to the problem” without mentioning that the elegant solution was found only after 13 less-elegant solutions. The smartest person in the world might have been able to find the elegant solution the first time, but most of us aren’t that smart.

There’s even a name for this: hindsight bias. That means you are much more likely to see how somebody thinks they should have solved the problem, not how they really solved the problem.

4  Checking

Let’s follow up on the idea of checking the work, as mentioned in item 4.

4.1  In General

28.
Design the work from the outset to be readily checkable.

29.
As a means to this end, leave breadcrumbs (or whatever) to mark your path. Very often, the solution requires a large number of steps. Document things in enough detail that each step can be checked locally, independently of the others. This makes checking vastly more efficient and vastly more reliable, compared to a situation where the only way to check the solution is to re-do the whole thing from scratch.

Solving a hard problem often includes a process that is equivalent to searching a maze. Mathematically this corresponds to enumerating the corners of a high-dimensional hypercube. In computer science, it is called walking the tree. In any case, when you return to a given node, it is important to know what’s been tried before; otherwise you risk gross inefficiency, up to and including infinte loops. Therefore you need to leave bread crumbs to mark the path.

This is super-extra important in conjunction with item 32.

An example of a relatively simple problem that rewards systematic tree-walking is Cheryl’s Birthday. A more elaborate example is Who Owns the Fish (reference 12).

cheryls-birthday
Figure 1: Logic Puzzle : Cheryl’s Birthday

30.
If you solved the problem in a flash of divine inspiration, go back and figure out a way it could have been solved methodically, step by step. This is necessary so that mere mortals can check your work. Also, you may discover that the inspired solution is only part of a larger solution-set.

Also: Inspiration is a way of obtaining an answer very quickly ... but beware that all too often, it is a way of obtaining a wrong answer very quickly.

31.
Have somebody else check your work.

32.
If you can’t find any individual with exactly the right skills to check the work, put together a team of specialists, each of whom can handle some part of the job.

33.
Check earlier rather than later.

Measure twice,
cut once.
     

Publish in haste;
repent at leisure.
     

34.
If there are more than one or two items, use a checklist to make sure that every part of the job gets done.

4.2  Checking a Number

35.
Check the order of magnitude. If the height of the tree comes out to be 10 km, you’ve done something wrong.

36.
Check the dimensions. For details, see reference 13. Also compare item 46.

37.
There are other things that are technically not dimensions, but can be checked in the same manner. An example is vector character. If the answer is supposed to be a vector, and you’ve got a scalar, you’ve done something wrong.

38.
Apply number-theoretic checks, such as casting out nines.

4.3  Checking a Formula

39.
Check a few typical cases numerically. Apply the checks discussed in section 4.2, such as magnitude, dimensions, et cetera.

40.
Check a few boundary cases (up against one constraint)

41.
Check a few edge cases (up against two constraints simultaneously)

42.
Check a few corner cases (up against multiple constraints simultaneously)

43.
Check a few symmetrical cases: Maybe the triangle is isosceles. Maybe the elliptical orbit is circular.

44.
Check a few simple and/or familiar special cases: Maybe the triangle is a 30-60-90 right triangle.

45.
Check the results against the various relevant symmetries, conservation laws, sum rules, et cetera.

46.
Check the scaling behavior. This is a more general and more powerful extension of item 36. For details, see reference 14.

47.
Check that your answer depends on the givens in the proper way.

For example, if you prove that a2 + b2 = c2 for «all» triangles, you’ve proved something that isn’t true. The statement should be restricted to right triangles in the Euclidean plane ... and your proofe of the statement should be clearly dependent on these restrictions.

It’s bad luck to prove things that aren’t true.
     

4.4  Checking the solution to a problem

48.
Solve the problem in more than one way. Any problem worth solving is worth solving twice. For example, if a physics problem can be solved in terms of forces, solve it again in terms of energy, i.e. the principle of virtual work. If you can apply Newton’s third law in terms of equal-and-opposite forces, apply it again in terms of momentum conservation.

Maybe for an easy, non-tricky problem you can get away with solving it only once, but if you are dealing exclusively with easy, non-tricky problems you wouldn’t be reading this document.

This is of course related to item item 4: check the work.

49.
Check that you have found the entire solution set. Sometimes simple-looking questions have multiple answers, some of which may be very much less obvious than others.

Example: South-West-North equilateral triangle on the surface of a sphere.

5  Formalism, Algorithms, et cetera

50.
As mentioned in item 11, formalism can be very powerful. Learn to use the formalism.

Formalism is a tool. Any tool can be abused ... but that doesn’t mean you are obliged to abuse it. Powerful tools, when properly used, make it easier (not harder) to understand what’s going on. For example: Consider the following contrast:

Maxwell’s paper on electrodynamics is almost unreadable. That’s because he didn’t use vector formalism. Vectors were not invented until many years later.   Nowadays, we write the Maxwell equations in terms of vectors ... or, preferably, bivectors.

  When you see the vector equation, you need to know the vector formalism to make sense of it.

Without vectors, you are left with a bunch of dry mathematical equations, and figuring out what they mean physically is exceedingly difficult.   Using vector ideas, you can visualize what’s going on. The vectors have a physical significance.

If you try to do without vectors, you wind up with so many equations it’s difficult to keep track of them.   The vector equations are quite compact. You can look at them all at once.

So, when trying to solve an electrodynamics problem, there are three possibilities:

  1. The best option is to learn to use the formalism properly, and use it to solve the problem.
  2. If you try to use the formalism without fully understanding it, you might get the right answer, or you might not. You might just generate a big fancy mess.
  3. If you try to solve the problem without using the formalism at all, you’re just wasting your time. Learning the formalism is far easier than trying to solve the problem without it.

Here’s another example that supports the same conclusion:

If you learn to play the piano, in the sense of hitting hitting the right notes, it does not automatically bestow artistry or a “feel” for the music.   Learning to play the piano does not prevent you from acquiring artistry and musical feeling, and may even help.

After all, you could program an unfeeling robot to hit the right notes.   Just because you “could” play like a robot doesn’t mean you are obliged to.

In all cases, remember the proverb:

It is a poor workman who blames his tools.
     

I say learn to use the tool, and then do something sensible with it.

6  Practice

51.
There is a saying that «practice makes perfect» but that is quite wrong. The fact is, practice makes permanent. Practicing the wrong things is worse than nothing.

Practice makes permanent.
     

7  References

1.
TeachMath
“Mathematical Habits of Mind”
http://honeymath.blogspot.com/2012/05/mathematical-habits-of-mind.html
2.
George Pólya,
How to Solve It: A New Aspect of Mathematical Method
1st edition (1945), 2nd edition (1957), Princeton University Press
3.
Timothy Gowers,
“Proving the fundamental theorem of arithmetic”
https://gowers.wordpress.com/2011/11/18/proving-the-fundamental-theorem-of-arithmetic/
4.
John Denker,
“Learning, Remembering, and Thinking”
www.av8n.com/physics/thinking.htm
5.
John Denker,
“Principles of Teaching and Learning”
www.av8n.com/physics/pedagogy.htm
6.
John Denker,
“Suggestions for Writing Good Software”
www.av8n.com/computer/htm/good-software.htm
7.
Wikipedia article: “Diagram”
https://en.wikipedia.org/wiki/Diagram#Specific_diagram_types
8.
John Denker,
“How to Deal with Ill-Posed Questions”
www.av8n.com/physics/ill-posed.htm
9.
John Denker,
“Perspectives on Johnson Noise”
www.av8n.com/physics/johnson-noise.htm
10.
John Denker,
“Modeling a Collision with Successive Refinement”
www.av8n.com/physics/collision-model-refine.htm
11.
John Denker,
“Hints on How to Do Math”
www.av8n.com/physics/math-hints.htm
12.
Puzzle: “Who Owns the Fish?”
www.av8n.com/physics/who-owns-the-fish.htm
13.
John Denker
“Dimensional Analysis”
www.av8n.com/physics/dimensional-analysis.htm
14.
John Denker,
“Scaling Laws”
www.av8n.com/physics/scaling.htm

1
Specifically: They had data in June 1963, with more and better data over the next year. Contact with the Big Bang community did not begin until March 1965.
[Contents]
Copyright © 2014 jsd