How to Study Differential Equations: Classify First

Study differential equations by drilling classification first. Almost every method in the course works on one shape of equation and fails on the rest, so the marks turn on naming the type before you solve anything. Build a fixed triage order, run it on unlabelled equations, and practise that naming step on its own.

The two courses that usually come before it ask for something else. Calculus rewards the volume of problems you work by hand and linear algebra rewards arguments built out of stated properties, which our guides to how to study calculus and how to study linear algebra set out in full. A differential equations course hands you a dozen named recipes and then stops telling you which one a question wants.

Why does a differential equations course feel like a cookbook?

Because the syllabus really is a list of named types with a procedure attached to each. The Lamar University differential equations notes, the free reference Paul Dawkins publishes from the courses he taught there, open their first order chapter by naming the contents outright: “In this chapter we will look at several of the standard solution methods for first order differential equations including linear, separable, exact and Bernoulli differential equations.” Later chapters add the characteristic equation for constant coefficient second order equations, undetermined coefficients, variation of parameters, series solutions and Laplace transforms. Each entry is a recipe, and the cookbook impression a student forms in week three reads the contents page accurately.

The published expectations read the same way. The MIT OpenCourseWare syllabus for 18.03 Differential Equations prints a list it calls the Ten Essential Skills, introduced with the line “Students should strive for personal mastery over the following skills.” The entries are procedures. Skill two is “Solve a first order linear ODE by the method of integrating factors or variation of parameter.” Skill four is “Solve a constant coefficient second order linear initial value problem with driving term exponential times polynomial.” The syllabus adds that the list “is widely disseminated among the faculty teaching courses listing 18.03 as a prerequisite”, and that at the time of that Spring 2010 version, 140 courses at MIT listed 18.03 as a prerequisite or a corequisite.

Here is where the impression stops being useful. A cookbook names the recipe you are cooking, and so does an exercise set, because the heading above it names the chapter. A comfortable evening of practice can therefore leave one whole skill untrained. On the exam the equation arrives bare, and choosing wrongly there produces no partial credit: an integrating factor built for an equation that was never linear yields a page of confident working and no marks at all.

What do you check first when you meet an unfamiliar equation?

Run a short sequence of cheap tests and stop at the first one that fits. Every test below is answered by looking at the shape of the equation, so the whole triage costs less than the first line of a wrong solution.

Two things about that order are worth stating plainly. First, the families overlap. A linear equation with nothing on the right hand side always separates, so an equation can qualify under two headings at once, and the triage is picking the cheaper route rather than the only legal one. Second, the order itself is a convention you adopt, and some courses check linearity before separability. Fix one order, write it inside the cover of the notebook, and run the same sequence every time until it costs no thought.

Every method carries a precondition on the form

The reason classification can be drilled at all is that each recipe states, in advance, the shape it needs. The Lamar notes are blunt about it in the linear section: “In order to solve a linear first order differential equation we MUST start with the differential equation in the form shown below.” That form is dy/dx + p(x)y = g(x). Only once the equation sits in it does the integrating factor mean anything.

The mechanics are worth holding in your head rather than memorising as a formula. Multiply through by a function mu(x) chosen so that the derivative of mu equals mu times p. The left hand side then collapses into the derivative of the product mu times y, by the ordinary product rule, and the whole equation can be integrated in one step. Solving the condition on mu gives mu(x) as e raised to the integral of p(x) dx.

That last row hides the single worst vocabulary trap in the subject. The word homogeneous carries two unrelated meanings in one course. A first order equation is called homogeneous when it can be written as dy/dx = F(y/x), which is the substitution case above. A linear equation of any order is called homogeneous when the right hand side is zero, and at second order with constant coefficients that is the case the characteristic equation solves. Nothing connects the two uses except the word. Write both meanings on one card early, because a student who has conflated them will reach for a substitution in a second order chapter and lose the question before starting it.

The exactness test earns its own note, since it is the clearest illustration of why classification pays. The Lamar notes introduce it by asking for exactly that saving: it would be useful, they write, to have “some simple test that we could use before even starting to see if a differential equation is exact or not”, because without it you might spend the session hunting a function that does not exist. Comparing two partial derivatives takes under a minute and rules the method in or out before you commit to it.

Second order asks a different question from first order

Once the order reads two, the triage changes shape. For a linear equation with constant coefficients and zero on the right hand side, substituting an exponential in the unknown collapses the whole problem into a quadratic, the characteristic equation ar^2 + br + c = 0. Three cases follow from the roots, and they are worth learning as a set of three rather than as three separate weeks: two real distinct roots, one repeated real root, and a complex conjugate pair, each producing its own standard form of general solution.

With something on the right hand side, the answer splits into the general solution of the homogeneous version plus one particular solution, and the classification question becomes which method supplies that particular solution. Undetermined coefficients is a guess at the form of the answer with the numbers left open, and it carries two preconditions the Lamar notes state directly: it “will only work for a fairly small class of functions” on the right hand side, and “it is generally only useful for constant coefficient differential equations”. Variation of parameters is described there as “a much more general method that can be used in many more cases”, at the cost of needing the homogeneous solution in hand first and of two integrals that may not come out.

Laplace transforms then sit slightly apart from all of this. They are aimed at initial value problems, and they earn their place on forcing terms that switch on partway through or arrive as an impulse, where the other methods become painful. So the second order branch of the triage runs its own short sequence. Ask whether the coefficients are constant, then whether the right hand side is zero, then which particular solution method the shape of that right hand side allows.

Practise the triage as its own exercise

Naming and solving are separate skills, and only one of them gets practised by working through a chapter in order. Build the other deliberately. Copy forty equations from across the term into one list, strip every heading, shuffle them, and then answer only the question of which method applies. No working, no integration, just a name and the one-line reason that licensed it. Forty equations takes about ten minutes and will surface confusions that a week of solving never touched.

Mixing is the whole mechanism here, and the general case for it is set out in our guide to the interleaving study method. Two details matter for this subject in particular. Draw from chapters you have already finished as well as the current one, because the exam will, and it will never tell you which week a question came from. Keep a short tally of the pairs you confuse, since confusions cluster rather than scatter, and a tally turns a vague feeling of struggling into two or three specific distinctions you can fix in an afternoon.

When a solve does go wrong, sort the cause before you revise anything. A misnamed equation needs more triage practice. A correctly named equation that fell apart in the integration needs calculus work instead, and that is the commonest cause by some distance. Systems of first order equations bring in a third cause, since their solutions run through eigenvalues and eigenvectors, so the eigenvalue material in the linear algebra guide above is the revision that helps there. Students meeting this course inside an engineering degree will find the wider toolkit in our roundup of the best AI study tools for engineering students.

How do you revise for a differential equations exam?

Put method cards ahead of the shuffled list, then put full solves behind it. A method card holds three things and nothing else: the precondition that licenses the method, the first line of working, and the shape of the answer it produces. Cover the card, say the precondition aloud, and check. The difference between retrieving a method and rereading a page of notes is the subject of active recall versus spaced repetition, and it applies with unusual force to a list of recipes that all look similar on paper. Once the cards are solid, the classification list gets timed, and the full solves come last, drawn from the same shuffled pile so the page never tells you the chapter.

Spread all three across the weeks rather than stacking them into the last one. A course built on a dozen similar procedures is exactly the material that blurs together when it is learned in one block, and the distinctions you want on exam day are the ones you have rehearsed while they had begun to fade.

Using GeniusPal to drill the naming step

The naming pass has an awkward property as homework. It needs a pile of unlabelled questions, and the material you own arrives sorted into chapters with the answer printed above the exercises. Building the shuffled pile by hand costs half an evening of typing, and it is the half that teaches you nothing.

Retrieval practice is the right tool for that pile. Dunlosky, Rawson, Marsh, Nathan and Willingham reviewed ten common study techniques in the 2013 review of learning techniques and gave their two high utility ratings to practice testing and distributed practice, because “they benefit learners of different ages and abilities”. Asking yourself which method an equation wants, then checking the answer, is practice testing aimed straight at the step the exam charges for.

GeniusPal builds that pile from a file you already have. Hand it a lecture handout, a problem sheet, a slide deck or a chapter, as a PDF, a Word or PowerPoint file, plain text, Markdown or CSV, anything up to 10 MB, or give it the address of a web page. Back comes a set of questions with answers drawn from that document, phrased the way your own course phrases things.

The plans decide how far that stretches. On Free the account gets 2 generations in total, which never refill, each producing 10 questions that can be sat twice, and every quiz there is multiple choice. Student costs $14.99 a month, raises the allowance to 100 generations a month and 30 questions a set, and unlocks the same questions as flashcards and as written active recall. Quiz, flashcards and written recall are the three modes in total. One thing stays free at every tier: a daily review of as many as 20 questions, assembled across all the sets in your account, leading with whatever you answered wrongly.

Then the limits, which bite harder in this subject than in most. GeniusPal does no mathematics itself: every integral, characteristic root and marked line of working stays yours to produce. It works from extracted text, so a photographed or scanned exercise sheet carrying no text layer never reaches a set. It also writes questions from the file in front of it, which means the pile is only ever as mixed as the material you hand over. One chapter in gives you one chapter of questions, sorted the way that chapter sorted them. A file covering the whole term gives the mixing somewhere to come from.

So fix the order of your checks this week, write it inside the cover of the notebook, and run it on every equation you meet before the first line of working. The textbook holds every method you will need and will go on holding them. The habit of naming the equation before you start is the part you build yourself.