Finite and Infinite Calculus — A Friendly Introduction
So you're comfortable with algebra but you've never taken calculus. Perfect — that's exactly the right starting point for this post. We're going to look at two parallel mathematical systems: the calculus you'd find in any university textbook, and a lesser-known twin called finite calculus that works in discrete steps instead of smooth curves. They mirror each other so closely it's almost eerie.
Don't worry if some of it feels unfamiliar at first. Every new symbol gets explained when it shows up, and we'll build everything up from things you already know.
First, a Quick Word on What Calculus Actually Is
At its heart, calculus is the mathematics of change. If you've ever asked "how fast is this growing?" or "how much total area is under this curve?", those are calculus questions.
The key tool is called an operator. An operator is just a machine that takes a function as input and spits out a different function as output. You feed in \(f\), and out comes something new. That's it. Both calculus systems in this post are built around one central operator each.
A function, if you need a refresher, is just a rule that turns one number into another. For example, \(f(x) = x^2\) turns 3 into 9, turns 5 into 25, and so on.
The Two Operators
Ordinary (infinite) calculus is built on the derivative operator \(D\). It measures the instantaneous rate of change — how fast a function is changing at a single exact point. The formal definition is:
$$Df(x) = \lim_{h \to 0} \frac{f(x+h) - f(x)}{h}$$All that's saying is: look at how much \(f\) changes when you nudge \(x\) by a tiny amount \(h\), divide by \(h\) to get a rate, and then see what happens as that nudge shrinks toward zero. The result is the slope of the curve at that exact point.
Finite calculus is built on the difference operator \(\Delta\) (the Greek capital letter delta, which conventionally stands for "change"). Instead of shrinking \(h\) toward zero, we just fix \(h = 1\) — a single unit step. So:
$$\Delta f(x) = f(x+1) - f(x)$$This is much simpler! It just asks: what's the difference between consecutive values of \(f\)? For example, if \(f(x) = x^2\), then \(\Delta f(3) = 4^2 - 3^2 = 16 - 9 = 7\). That's it.
Both \(D\) and \(\Delta\) are called operators because they take a function in and hand a new function back out.
The Power Rule for \(D\) — The First Big Result
The power rule is one of the first things you'd learn in any calculus class. It tells you what happens when \(D\) acts on a power of \(x\), like \(x^2\) or \(x^5\).
The rule is beautifully simple:
$$D(x^m) = mx^{m-1}$$Translation: bring the exponent down in front as a multiplier, then subtract 1 from the exponent. That's literally all you do.
| Start with \(f(x)\) | After applying \(D\) | What happened |
|---|---|---|
| \(x^1\) (just \(x\)) | \(1 \cdot x^0 = 1\) | Brought down the 1, reduced exponent to 0. Since \(x^0 = 1\), we just get 1. |
| \(x^2\) | \(2x\) | Brought down the 2, reduced exponent from 2 to 1. |
| \(x^3\) | \(3x^2\) | Brought down the 3, reduced exponent from 3 to 2. |
| \(x^5\) | \(5x^4\) | Brought down the 5, reduced exponent from 5 to 4. |
| \(x^0 = 1\) (a constant) | \(0\) | A constant never changes, so its rate of change is zero. |
The rule even works for fractions and negative exponents, which is one of the reasons it's so powerful:
$$D(x^{1/2}) = \tfrac{1}{2}x^{-1/2} = \frac{1}{2\sqrt{x}}, \qquad D(x^{-1}) = -x^{-2} = -\frac{1}{x^2}, \qquad D(x^{\pi}) = \pi x^{\pi-1}$$Why Does This Work?
The short version: if you expand \((x+h)^m\) using the binomial theorem (the rule for multiplying out brackets raised to a power), the first two terms are \(x^m + mx^{m-1}h\). When you subtract \(x^m\) and divide by \(h\), you get \(mx^{m-1}\) plus leftover terms that all still contain \(h\). As \(h\) shrinks to zero, those leftovers vanish. What remains is exactly \(mx^{m-1}\).
Applying \(D\) More Than Once
Because \(D\) turns a function into another function, you can apply it again. Each application peels off another factor:
$$D^2(x^m) = m(m-1)x^{m-2}, \qquad D^k(x^m) = m(m-1)(m-2)\cdots(m-k+1)\,x^{m-k}$$That product \(m(m-1)(m-2)\cdots\) — where you keep multiplying by one less each time — is called a falling factorial. Keep this in mind; it shows up again very soon.
So What Happens When \(\Delta\) Meets Ordinary Powers?
Given how clean the power rule is for \(D\), you'd hope \(\Delta\) would be just as tidy. Spoiler: it's not. Let's try \(x^3\):
$$\Delta(x^3) = (x+1)^3 - x^3 = 3x^2 + 3x + 1$$We wanted something clean like \(3x^2\), but we got two extra terms tagging along. The problem is that ordinary powers like \(x^3\) and discrete unit steps don't fit together cleanly. Raising \((x+1)\) to a power generates cross-terms that don't cancel out.
This is a real problem if you want finite calculus to work elegantly. The fix is surprisingly neat: we don't change the operator, we change the powers.
Meet the Factorial Powers — The Right Tool for the Job
Instead of ordinary powers like \(x^3 = x \cdot x \cdot x\), finite calculus uses factorial powers, where the factors count downward (or upward) instead of repeating the same value.
Falling Factorial Powers
The falling factorial \(x^{\underline{m}}\) is read "x to the m falling." You multiply \(m\) factors together, starting at \(x\) and stepping down by 1 each time:
$$x^{\underline{m}} = \overbrace{x(x-1)(x-2)\cdots(x-m+1)}^{m \text{ factors}}$$Here are some concrete examples for \(x = 5\):
| Notation | Expanded | Value at \(x=5\) |
|---|---|---|
| \(x^{\underline{1}}\) | \(x\) | \(5\) |
| \(x^{\underline{2}}\) | \(x(x-1)\) | \(5 \cdot 4 = 20\) |
| \(x^{\underline{3}}\) | \(x(x-1)(x-2)\) | \(5 \cdot 4 \cdot 3 = 60\) |
| \(x^{\underline{0}}\) | (empty product) | \(1\) by convention |
Compare \(x^3 = 5 \cdot 5 \cdot 5 = 125\) with \(x^{\underline{3}} = 5 \cdot 4 \cdot 3 = 60\). Similar idea, different result. The underline under the exponent is your reminder that the factors step down.
Rising Factorial Powers
There's also a rising version, \(x^{\overline{m}}\), read "x to the m rising," where the factors step up:
$$x^{\overline{m}} = \overbrace{x(x+1)(x+2)\cdots(x+m-1)}^{m \text{ factors}}$$For example, \(5^{\overline{3}} = 5 \cdot 6 \cdot 7 = 210\). The overline is your reminder that the factors go up.
Connection to Factorials
You've probably seen the factorial before: \(n! = n \times (n-1) \times \cdots \times 2 \times 1\). For example, \(5! = 120\). These factorial powers are a generalization of that idea. In fact:
$$n! = n^{\underline{n}} = 1^{\overline{n}}$$In words: \(n\) factorial equals \(n\) to the \(n\) falling, which also equals 1 to the \(n\) rising. You can verify the last one: \(1^{\overline{4}} = 1 \cdot 2 \cdot 3 \cdot 4 = 24 = 4!\). ✓
The Power Rule for \(\Delta\) — The "Basic Factorial Fact"
Here's the payoff for inventing factorial powers. Watch what happens when \(\Delta\) acts on a falling factorial:
$$\Delta(x^{\underline{m}}) = (x+1)^{\underline{m}} - x^{\underline{m}}$$Let's work it out for \(m = 3\) to see the pattern. We need \((x+1)^{\underline{3}} - x^{\underline{3}}\):
$$\underbrace{(x+1) \cdot x \cdot (x-1)}_{\text{start at }x+1,\text{ step down}} - \underbrace{x \cdot (x-1) \cdot (x-2)}_{\text{start at }x,\text{ step down}}$$Factor out \(x(x-1)\), which appears in both terms:
$$= x(x-1)\,\bigl[(x+1) - (x-2)\bigr] = x(x-1) \cdot 3 = 3x^{\underline{2}}$$In general, the result is always the same clean pattern:
$$\boxed{\Delta(x^{\underline{m}}) = m\,x^{\underline{m-1}}}$$This is called the basic factorial fact, and it's the finite calculus analog of \(D(x^m) = mx^{m-1}\). Put them side by side:
| System | Power rule | Example |
|---|---|---|
| Infinite calculus | \(D(x^m) = mx^{m-1}\) | \(D(x^3) = 3x^2\) |
| Finite calculus | \(\Delta(x^{\underline{m}}) = mx^{\underline{m-1}}\) | \(\Delta(x^{\underline{3}}) = 3x^{\underline{2}}\) |
Identical structure. The only difference is the type of power being used.
Going Backwards: Anti-Derivatives and Anti-Differences
In ordinary calculus, you can run the derivative machine in reverse. If you know the output, you can find the original function. This reverse process is called the anti-derivative (or integral), written \(\int\).
The relationship is captured by the Fundamental Theorem of Calculus, which says:
$$g(x) = Df(x) \quad\Longleftrightarrow\quad \int g(x)\,dx = f(x) + C$$The \(C\) is a constant — because the derivative of any constant is zero, you can always add a constant to the answer and it's still valid. The \(dx\) is just a notation convention that says "we're integrating with respect to \(x\)."
Finite calculus has an exact analog. The reverse of \(\Delta\) is called the anti-difference (or indefinite sum), written \(\sum g(x)\,\delta x\):
$$g(x) = \Delta f(x) \quad\Longleftrightarrow\quad \sum g(x)\,\delta x = f(x) + C$$Notice the lowercase \(\delta\) in \(\delta x\) — it parallels the lowercase \(d\) in \(dx\), just as uppercase \(\Delta\) parallels uppercase \(D\). The constant \(C\) here is any function that repeats every time you move one unit — like \(\sin(2\pi x)\), which has the same value at every integer. At integer values of \(x\), \(C\) is just a plain constant.
Definite Integrals and Definite Sums
You can also do integrals between two specific values, giving a single number as the result instead of a function. If \(g(x) = Df(x)\):
$$\int_a^b g(x)\,dx = f(x)\Big|_a^b = f(b) - f(a)$$The notation \(f(x)\Big|_a^b\) means "evaluate \(f\) at \(b\), then subtract \(f\) at \(a\)." So the definite integral is just the difference in \(f\)'s values at the two endpoints.
Finite calculus has an exact parallel. If \(g(x) = \Delta f(x)\):
$$\sum_a^b g(x)\,\delta x = f(x)\Big|_a^b = f(b) - f(a)$$But What Does \(\sum_a^b g(x)\,\delta x\) Actually Add Up?
This is worth unpacking. It turns out that for integers \(b \ge a\):
$$\sum_a^b g(x)\,\delta x = \sum_{k=a}^{b-1} g(k) = g(a) + g(a+1) + g(a+2) + \cdots + g(b-1)$$The key detail: the upper limit \(b\) is excluded. You sum from \(a\) up to but not including \(b\). This is different from ordinary integration, where both endpoints contribute.
Why does this work out so cleanly? Write out the sum in full:
$$\bigl(f(a+1)-f(a)\bigr) + \bigl(f(a+2)-f(a+1)\bigr) + \cdots + \bigl(f(b)-f(b-1)\bigr)$$Look at what cancels: \(f(a+1)\) appears as a positive term in the first bracket and a negative term in the second. Same for \(f(a+2)\), \(f(a+3)\), and so on — everything in the middle cancels out. All that survives is \(f(b) - f(a)\). These are called telescoping sums, named after a collapsible telescope — the middle sections fold away and only the outermost and innermost pieces remain.
Just like integrals, you can reverse the limits (which negates the result) or split the sum over an intermediate point:
$$\sum_a^b g(x)\,\delta x = -\sum_b^a g(x)\,\delta x, \qquad \sum_a^b + \sum_b^c = \sum_a^c$$Summing Falling Powers — Where It Gets Really Useful
The basic factorial fact \(\Delta(x^{\underline{m}}) = mx^{\underline{m-1}}\) means that going in reverse, the anti-difference of \(x^{\underline{m}}\) is \(\frac{x^{\underline{m+1}}}{m+1}\). That gives us a beautiful summation formula:
$$\sum_{0 \le k \lt n} k^{\underline{m}} = \frac{n^{\underline{m+1}}}{m+1}, \quad \text{for whole numbers } m \text{ and } n$$Compare this side by side with the corresponding integral:
$$\int_0^n x^m\,dx = \frac{n^{m+1}}{m+1}$$Practically identical — ordinary powers replaced by falling powers, sum replaced by integral. When \(m = 1\) (since \(k^{\underline{1}} = k\)):
$$\sum_{0 \le k \lt n} k = \frac{n^{\underline{2}}}{2} = \frac{n(n-1)}{2}$$That's a closed form for the sum \(0 + 1 + 2 + \cdots + (n-1)\). Check it: for \(n = 5\), we get \(\frac{5 \cdot 4}{2} = 10\), and \(0+1+2+3+4 = 10\). ✓
Summing Ordinary Powers Using Falling Powers
Ordinary powers like \(k^2\) can be converted into falling powers. For squares, the identity is:
$$k^2 = k^{\underline{2}} + k^{\underline{1}}$$You can verify this: \(k^{\underline{2}} + k^{\underline{1}} = k(k-1) + k = k^2 - k + k = k^2\). ✓
Using this identity, we can sum all the squares from \(0\) to \(n-1\):
$$\sum_{0 \le k \lt n} k^2 = \frac{n^{\underline{3}}}{3} + \frac{n^{\underline{2}}}{2} = \frac{1}{3}n\!\left(n-\tfrac{1}{2}\right)\!(n-1)$$Replace \(n\) with \(n+1\) to get the sum from \(0\) to \(n\), which gives the classic formula:
$$\sum_{0 \le k \le n} k^2 = \frac{n(n+1)(2n+1)}{6} \checkmark$$For cubes, the identity is \(k^3 = k^{\underline{3}} + 3k^{\underline{2}} + k^{\underline{1}}\), giving:
$$\sum_{a \le k \lt b} k^3 = \left(\frac{k^{\underline{4}}}{4} + k^{\underline{3}} + \frac{k^{\underline{2}}}{2}\right)\Bigg|_a^b$$Negative Falling Powers — Extending the Pattern
So far we've only looked at \(x^{\underline{m}}\) when \(m\) is a positive whole number. What about negative values of \(m\)?
Look at the sequence as \(m\) decreases:
$$x^{\underline{3}} = x(x-1)(x-2), \quad x^{\underline{2}} = x(x-1), \quad x^{\underline{1}} = x, \quad x^{\underline{0}} = 1$$Each step divides by the last factor added: going from \(x^{\underline{3}}\) to \(x^{\underline{2}}\), you divide by \((x-2)\). Going from \(x^{\underline{2}}\) to \(x^{\underline{1}}\), you divide by \((x-1)\). Going from \(x^{\underline{1}}\) to \(x^{\underline{0}}\), you divide by \(x\). So the next step — going from \(x^{\underline{0}}\) to \(x^{\underline{-1}}\) — divides by \((x+1)\):
$$x^{\underline{-1}} = \frac{1}{x+1}, \qquad x^{\underline{-2}} = \frac{1}{(x+1)(x+2)}, \qquad x^{\underline{-3}} = \frac{1}{(x+1)(x+2)(x+3)}$$In general, for a positive whole number \(m\):
$$x^{\underline{-m}} = \frac{1}{(x+1)(x+2)\cdots(x+m)}$$Why This Definition Is Forced on Us
For ordinary powers, there's a law: \(x^{m+n} = x^m \cdot x^n\). Falling powers have their own version:
$$x^{\underline{m+n}} = x^{\underline{m}} \cdot (x-m)^{\underline{n}}$$Notice the shift: the second factor starts at \(x - m\) rather than \(x\), because the first factor already "used up" the values from \(x\) down to \(x-m+1\). Let's verify this with \(m = 2\) and \(n = -3\):
$$x^{\underline{2-3}} = x^{\underline{2}} \cdot (x-2)^{\underline{-3}} = x(x-1) \cdot \frac{1}{(x-1)x(x+1)} = \frac{1}{x+1} = x^{\underline{-1}} \checkmark$$If we had defined \(x^{\underline{-1}}\) as \(1/x\) instead of \(1/(x+1)\), this law would have broken. The definition we chose is the only one that keeps the law intact — and in mathematics, when you're extending a definition to new territory, preserving the existing rules is always the right guide.
The Basic Factorial Fact Still Works for Negative Powers
Let's check \(m = -2\):
$$ \begin{aligned} \Delta(x^{\underline{-2}}) &= \frac{1}{(x+2)(x+3)} - \frac{1}{(x+1)(x+2)} \\ &= \frac{(x+1)-(x+3)}{(x+1)(x+2)(x+3)} = \frac{-2}{(x+1)(x+2)(x+3)} = -2\,x^{\underline{-3}} \end{aligned} $$Sure enough, \(\Delta(x^{\underline{-2}}) = -2\,x^{\underline{-3}}\), exactly as the basic factorial fact predicts. So the summation law works for all \(m\) except one:
$$\sum_a^b x^{\underline{m}}\,\delta x = \frac{x^{\underline{m+1}}}{m+1}\Bigg|_a^b, \quad \text{for } m \ne -1$$The Special Case \(m = -1\): Harmonic Numbers and the Discrete Logarithm
In ordinary calculus, the case \(m = -1\) is special too:
$$\int x^{-1}\,dx = \ln x + C$$The natural logarithm \(\ln x\) is the anti-derivative of \(1/x\). It occupies a unique place — it doesn't fit the normal power rule because dividing by zero would appear in the formula.
Finite calculus has the same issue. When \(m = -1\), we need a function \(f(x)\) such that:
$$\Delta f(x) = x^{\underline{-1}} = \frac{1}{x+1}$$In plain English: we need a function whose consecutive differences give \(\frac{1}{x+1}\). The answer, for positive integer \(x\), is:
$$f(x) = 1 + \frac{1}{2} + \frac{1}{3} + \cdots + \frac{1}{x} = H_x$$This is the harmonic number \(H_x\) — the sum of the reciprocals of the first \(x\) whole numbers. You can check: \(H_{x+1} - H_x = \frac{1}{x+1}\). ✓
So harmonic numbers are the discrete analog of the natural logarithm. Just as \(\ln x\) appears constantly in continuous mathematics, \(H_x\) appears constantly in discrete mathematics — in the analysis of sorting algorithms, in probability, and in many counting problems. The complete summation law is:
$$\sum_a^b x^{\underline{m}}\,\delta x = \begin{cases} \dfrac{x^{\underline{m+1}}}{m+1}\Bigg|_a^b & \text{if } m \ne -1 \\[10pt] H_x\Big|_a^b & \text{if } m = -1 \end{cases}$$The Discrete Exponential: \(2^x\)
In ordinary calculus, \(e^x\) has a magical property: its derivative is itself. That is, \(De^x = e^x\). What is the discrete equivalent — a function whose difference is itself?
$$\Delta f(x) = f(x) \;\implies\; f(x+1) - f(x) = f(x) \;\implies\; f(x+1) = 2f(x)$$Each value is double the previous one. That's just \(f(x) = 2^x\). So \(2^x\) plays the role of \(e^x\) in finite calculus.
More generally, for any constant \(c \ne 1\):
$$\Delta(c^x) = c^{x+1} - c^x = (c-1)c^x$$The anti-difference of \(c^x\) is therefore \(\frac{c^x}{c-1}\). Attach limits and you get the geometric series formula for free:
$$\sum_{a \le k \lt b} c^k = \frac{c^b - c^a}{c-1}, \quad c \ne 1$$This is just the standard formula for a geometric series — finite calculus gives us a clean, systematic way to derive it.
A Handy Reference Table
Every function \(f\) with a known difference \(\Delta f = g\) gives us a summation formula for \(g\). Here are the most useful pairs:
| Anti-difference \(f\) (i.e., \(\sum g\)) | Difference \(g\) (i.e., \(\Delta f\)) |
|---|---|
| \(x^{\underline{0}} = 1\) | \(0\) |
| \(x^{\underline{1}} = x\) | \(1\) |
| \(x^{\underline{2}} = x(x-1)\) | \(2x^{\underline{1}} = 2x\) |
| \(x^{\underline{m+1}}/(m+1)\) | \(x^{\underline{m}}\) |
| \(H_x\) (harmonic number) | \(x^{\underline{-1}} = \frac{1}{x+1}\) |
| \(2^x\) | \(2^x\) |
| \(c^x/(c-1)\) | \(c^x\) |
| \(cf\) | \(c\,\Delta f\) (constants factor out) |
| \(f + g\) | \(\Delta f + \Delta g\) (sums split) |
Summation by Parts — The Discrete Integration by Parts
In ordinary calculus, there's a technique called integration by parts for handling integrals of products. It comes from the product rule for derivatives: \(D(uv) = u\,Dv + v\,Du\), rearranged to:
$$\int u\,Dv = uv - \int v\,Du$$The strategy: if \(\int u\,Dv\) is hard to compute, maybe \(\int v\,Du\) is easier. Finite calculus has an exact analog.
Start with the difference of a product \(u(x)v(x)\):
$$ \begin{aligned} \Delta(u(x)v(x)) &= u(x+1)v(x+1) - u(x)v(x) \\ &= u(x)\,\Delta v(x) + v(x+1)\,\Delta u(x) \end{aligned} $$The term \(v(x+1)\) is written more compactly using the shift operator \(E\), which just means "evaluate at \(x+1\)": \(Ef(x) = f(x+1)\). So the product rule for differences is:
$$\Delta(uv) = u\,\Delta v + Ev\,\Delta u$$Taking the anti-difference of both sides and rearranging gives summation by parts:
$$\sum u\,\Delta v = uv - \sum Ev\,\Delta u$$Example: Adding Up \(k \cdot 2^k\)
Let's compute \(\sum_{k=0}^{n} k\,2^k\) — the sum of \(0, 2, 8, 24, 64, \ldots\)
We set \(u(x) = x\) and \(\Delta v(x) = 2^x\). Then \(\Delta u(x) = 1\), \(v(x) = 2^x\), and \(Ev(x) = 2^{x+1}\). Plugging into summation by parts:
$$\sum x\,2^x\,\delta x = x\cdot 2^x - \sum 2^{x+1}\,\delta x = x\cdot 2^x - 2^{x+1} + C$$Attaching limits — and noting that a sum from \(k = 0\) to \(k = n\) in ordinary notation corresponds to the definite sum from 0 to \(n+1\) (because the upper limit is excluded):
$$ \begin{aligned} \sum_{k=0}^{n} k\,2^k &= \bigl(x\cdot 2^x - 2^{x+1}\bigr)\Big|_0^{n+1} \\ &= \bigl((n+1)2^{n+1} - 2^{n+2}\bigr) - (0 - 2) \\ &= (n-1)2^{n+1} + 2 \end{aligned} $$Check for \(n = 3\): \(0 + 2 + 8 + 24 = 34\), and \((3-1) \cdot 2^4 + 2 = 2 \cdot 16 + 2 = 34\). ✓
Example: Summing \(k \cdot H_k\)
Here's a trickier one: \(\sum_{0 \le k \lt n} k\,H_k\), where \(H_k\) is the \(k\)th harmonic number.
We set \(u(x) = H_x\) and \(\Delta v(x) = x^{\underline{1}}\). Then: \(\Delta u(x) = x^{\underline{-1}}\), \(v(x) = x^{\underline{2}}/2\), and \(Ev(x) = (x+1)^{\underline{2}}/2\).
$$ \begin{aligned} \sum x\,H_x\,\delta x &= \frac{x^{\underline{2}}}{2}\,H_x - \frac{1}{2}\sum (x+1)^{\underline{2}}\,x^{\underline{-1}}\,\delta x \end{aligned} $$The product \((x+1)^{\underline{2}} \cdot x^{\underline{-1}}\) simplifies using the law of exponents. Since \((x+1)^{\underline{2}} = (x+1)x\) and \(x^{\underline{-1}} = \frac{1}{x+1}\), the product is just \(x = x^{\underline{1}}\). So:
$$ \begin{aligned} &= \frac{x^{\underline{2}}}{2}\,H_x - \frac{1}{2}\sum x^{\underline{1}}\,\delta x \\ &= \frac{x^{\underline{2}}}{2}\,H_x - \frac{x^{\underline{2}}}{4} + C \end{aligned} $$Attaching limits:
$$\sum_{0 \le k \lt n} k\,H_k = \frac{n^{\underline{2}}}{2}\!\left(H_n - \frac{1}{2}\right)$$And as a bonus: we could also derive \(\sum_{0 \le k \lt n} H_k = nH_n - n\) the same way, instead of stumbling across it by lucky substitution.
Comments