One equation, two ways to step
Suppose and a current approximation is known at . Explicit Euler uses the slope at the old point: . Implicit Euler instead asks the unknown next value to satisfy
A formula involving an unknown is an equation to solve, not an instruction to substitute the old value. For nonlinear , an implicit step may require an iterative solver; existence, uniqueness and stopping error need separate attention.
The decay test
To compare the two updates we need a precise notion of "the numerical solution behaves like a decaying exact solution". The simplest test problem is with real, whose exact solutions tend to zero.
Definition 9.1.1
(Stability on the decay test equation)
Let a one-step method applied to with step have the form , where the multiplier depends only on the product . The method is stable at the step (for this ) if , so that for every starting value .
Explicit Euler has multiplier . Implicit Euler gives , hence .
Theorem 9.1.2
(Decay of the two Euler methods)
Let be real and .
- The implicit multiplier satisfies for every , so implicit Euler is stable at every step size.
- Explicit Euler is stable exactly when . If then and for all ; if and then .
Proof
(The multipliers and their size)
Since and we have , so and . This proves part 1.
For part 2, by induction from . Now means , that is ; with this is . At the multiplier is , so has constant magnitude. For we have , hence and grows without bound when . If both methods stay identically zero.
Proof X-ray
Where each hypothesis acts
The proof has one structural move: it replaces the whole numerical trajectory by a single number, , so the long-run question becomes the size of . The sign of is used twice: in the implicit case it makes the denominator exceed , and in the explicit case it makes the right half of hold automatically, leaving the left half, , as the genuine step-size restriction. The case is a genuine boundary, not a tie: the magnitudes neither decay nor grow. A single small-looking step, such as a first value of modest size, proves nothing about , because stability is a statement about every power of . This result is about the scalar test equation; it does not assert accuracy for any differential equation.
Worked comparison
Worked example 9.1.3
(A large step: stable but inaccurate)
Take , , and . The exact solution is . Explicit Euler has , so its first two values are and ; it oscillates and grows. Implicit Euler has , so the first two values are and . These decay, but the exact first value is , quite different from : the stable value is more than three times too large. Stability alone is not a guarantee of accuracy at a large step.
Worked example 9.1.4
(An implicit step that is a quadratic equation)
Take , , , with exact solution , so . Implicit Euler requires , that is , with roots . We select the root that continues the positive solution, ; the other root is a spurious solution of the algebraic equation. Explicit Euler gives instead. Both are poor at this large step, but the implicit value stays positive like the exact solution, at the price of solving an equation.
Worked example 9.1.5
(The boundary step size)
For the explicit threshold is . At and , and the explicit values are : no decay and no growth. The implicit values at the same step are , giving . Starting instead from , both methods give at every step, so the stability question is only visible for nonzero data.
Common mistake
(Putting the old value on the right of the implicit formula)
Writing and calling it implicit Euler returns the explicit multiplier in the worked comparison, not . In the implicit formula the unknown appears on both sides; collect it first, then divide by .
Counterexample mode
Dropping the sign of the parameter breaks the implicit multiplier
One might read the theorem as saying that implicit Euler always produces a bounded, sensible step. This fails without . Take , . With the implicit equation is , which has no solution. With it is , so and , although the exact solution grows and never changes sign. The decay-test conclusions hold only for the hypotheses , .
A reproducible experiment
The following MATLAB code starts from the same initial value, uses a fixed number of steps and compares both methods at the same time points. It uses the algebraic implicit formula for this linear equation, so no nonlinear solver is hidden.
lambda = -4; h = 0.6; N = 5;
t = (0:N)*h;
ye = zeros(1,N+1); yi = ye;
ye(1) = 1; yi(1) = 1;
for j = 1:N
ye(j+1) = (1+h*lambda)*ye(j);
yi(j+1) = yi(j)/(1-h*lambda);
end
exact = exp(lambda*t);
disp([t(:), ye(:), yi(:), exact(:)])
To assess convergence, repeat with smaller while holding the final time fixed and choosing an integer step count. State the time interval, method, step size and error measured against the exact solution. The table alone does not prove a general stability theorem.
Exercises and solutions
Quick check
(For and nonzero , find all positive explicit-Euler steps giving decay.)
Use the condition .
Solution · Answer
means .
Quick check
(For and , what happens to a nonzero explicit iterate?)
Compute .
Solution · Answer
, so the signs alternate and the magnitudes stay constant.
Quick check
(For , starting from with , compute the first implicit step.)
Use .
Solution · Answer
, so .
Quick check
(Is a stable result necessarily close to the exact solution?)
Recall the worked comparison.
Solution · Answer
No. In the worked comparison the implicit value is stable but far from the exact .