Riemann Console dot org

An open research record on the Riemann Hypothesis.

NAV READY · TYPE SHORTCUT · ENTER EXECUTES

What is analytic continuation?

Series
Explain
Summary
A beginner-first, full-length article explaining analytic continuation from the geometric series and complex-plane intuition through analytic rigidity, the identity theorem, overlapping power-series continuation, zeta, Gamma, branching, Riemann surfaces and the Riemann Hypothesis connection.
Math Level
GENERAL
Index Excerpt
Analytic continuation extends an analytic function beyond the region where its original representation converges; complex-analytic rigidity makes that continuation highly constrained and gives zeta access to the critical strip.

There is a mathematical move that sounds, at first, as though it should not be allowed.

Consider the infinite series

1+z+z2+z3+z4+⋯ .1+z+z^2+z^3+z^4+\cdots.

If

∣z∣<1,|z|<1,

the terms shrink quickly enough for the series to converge.

(The condition ∣z∣<1|z|<1 means that the complex number zz lies less than one unit from zero. On the complex plane, those points fill the inside of a circle of radius 1.)

And in that region, the whole infinite sum has a beautifully simple value:

1+z+z2+z3+⋯=11−z.1+z+z^2+z^3+\cdots=\frac{1}{1-z}.

So far, so familiar.

But now choose

z=2.z=2.

The original series becomes

1+2+4+8+16+⋯ ,1+2+4+8+16+\cdots,

which plainly does not converge.

Yet the expression on the other side gives

11−2=−1.\frac{1}{1-2}=-1.

What has happened?

Have mathematicians somehow decided that

1+2+4+8+⋯=−1?1+2+4+8+\cdots=-1?

No.

Something subtler has happened.

The infinite series and the function it describes are not quite the same thing.

The series is one representation of the function, valid only in part of the complex plane.

(A complex number has a real part and an imaginary part, so instead of placing complex numbers on a one-dimensional number line we place them on a two-dimensional complex plane. Moving left or right changes the real part; moving up or down changes the imaginary part.)

The function can sometimes continue beyond the place where that particular representation stops working.

That process is called analytic continuation.

(Analytic continuation does not make a divergent series converge. It extends the analytic function that the series originally represented. The distinction between those two statements is fundamental.)

The formula can stop before the function does

Suppose we begin with

f(z)=1+z+z2+z3+⋯ .f(z)=1+z+z^2+z^3+\cdots.

Inside the unit circle,

∣z∣<1,|z|<1,

we can prove that

f(z)=11−z.f(z)=\frac{1}{1-z}.

The power series has a boundary.

The expression

11−z\frac{1}{1-z}

does not have the same boundary.

It makes perfectly good sense for

z=2,z=10,z=−7,z=i,z=2,\qquad z=10,\qquad z=-7,\qquad z=i,

and almost every other complex number.

There is just one obvious obstruction:

z=1,z=1,

where the denominator becomes zero.

So the same function that first appeared to live only inside ∣z∣<1|z|<1 can be recognised as part of a much larger object.

ASCII FIGURE // The formula stops before the function

A conceptual diagram showing a power-series formula valid inside a restricted region and the same analytic function continuing into a larger domain, except at a singular point.

A representation may have a smaller domain than the analytic function it represents.

This is the first conceptual step.

A formula is not necessarily the whole mathematical object.

The formula may be a window through which we first encounter it.

Analytic continuation asks whether the object visible through that window belongs to a larger analytic landscape.

Why call it analytic?

The word analytic is doing most of the work.

An ordinary function can behave almost however we want. We could define it one way for negative numbers, another way between zero and ten, and something completely different afterwards.

Even a smooth real-valued function on one interval does not usually determine what it must do everywhere else.

Complex analytic functions are very different.

They are extraordinarily rigid.

A complex function is called holomorphic, or complex analytic, in a region if it is complex-differentiable there.

That sounds like a small requirement.

It is not.

(For a real function, a derivative compares nearby values along one number line. In the complex plane we can approach a point from infinitely many directions. For a complex derivative to exist, all those directional approaches must fit together into one consistent value. That is a much stronger demand.)

Complex differentiability is so restrictive that an analytic function automatically has derivatives of every order and can locally be represented by a power series.

Around a point z0z_0, we can write

f(z)=a0+a1(z−z0)+a2(z−z0)2+⋯ .f(z)=a_0+a_1(z-z_0)+a_2(z-z_0)^2+\cdots.

Those coefficients are not independent decorations.

They encode the local behaviour of the function.

In a remarkable sense, enough local information can determine the function much farther away.

A function with very little freedom

One of the key results of complex analysis is the identity theorem.

In an informal version, it says this:

if two analytic functions agree on a sufficiently rich set of points inside a connected region, with an accumulation point in that region, then they are the same analytic function throughout the connected region where both are defined.

An open patch is more than enough.

Suppose two analytic functions agree at every point of some little open area. We cannot then alter one of them elsewhere, keep both analytic, and still claim that they are different continuations of the same original function.

The agreement propagates.

This is why analytic continuation is not mathematical improvisation.

If a continuation exists, it is tightly constrained by the function we already have.

(The identity theorem is stronger than saying that two formulas happen to match at a few sample points. Analytic functions can agree at infinitely many isolated points without being identical if those points have no accumulation point inside the domain. What matters is sufficiently dense local agreement.)

The original region does not merely supply some values.

It supplies analytic information.

And analytic information is remarkably difficult to change locally without changing the whole function.

How continuation actually moves

There is a useful way to picture the process.

Suppose a function is represented by a power series around one point.

That series converges inside some disc.

Choose another point inside that disc.

Around the new point, the same function has another power-series expansion.

That second series may converge into territory that the first one could not reach.

Then choose another point inside the second region.

And continue again.

The function can be carried across the complex plane by overlapping neighbourhoods.

ASCII FIGURE // Analytic continuation by overlapping regions

A chain of overlapping discs carries an analytic function from its original region into new territory while neighbouring descriptions agree on every overlap.

Analytic continuation can proceed step by step: each new local description agrees with the previous one where their domains overlap.

The overlaps matter.

They are the handshakes between neighbouring descriptions.

On an overlap, both formulas must describe the same analytic function. The identity theorem then prevents the continuation from wandering off into an unrelated choice.

(The radius of a particular power-series expansion is limited by the nearest obstruction to analyticity, often a singularity. Re-centering the series at another point can therefore let a new expansion reach territory that the first expansion could not.)

This is one reason complex analysis can feel surprising in its determinism.

Knowing the function here can constrain what the function must be there.

A map analogy, with a warning

There is a tempting analogy.

Imagine that we possess a detailed map of one region.

At its edge, another map overlaps it. Because the two maps share landmarks, roads and coordinates in the overlap, we can align them.

Then another overlapping map takes us farther.

And another farther still.

That is a useful picture of analytic continuation.

But the analogy has a limit.

Ordinary maps are created by surveying territory that already exists independently.

Analytic continuation is stronger.

The analytic structure already known in one region can determine what the continuation must be in the next.

We are not merely discovering an arbitrary neighbouring sheet.

The pieces are mathematically locked together.

(The overlapping-map picture explains the geometry of continuation, but not its rigidity. The identity theorem supplies that extra ingredient: compatible analytic pieces cannot disagree once their overlap has fixed them.)

The geometric series revisited

We can now return to

1+z+z2+z3+⋯ .1+z+z^2+z^3+\cdots.

Inside

∣z∣<1,|z|<1,

the geometric-series argument gives

1+z+z2+z3+⋯=11−z.1+z+z^2+z^3+\cdots=\frac{1}{1-z}.

The function on the right is analytic everywhere except at z=1z=1.

So

11−z\frac{1}{1-z}

provides an analytic continuation of the function originally represented by the series.

At z=2z=2, the continued function has the value −1-1.

But the original infinite series at z=2z=2 remains

1+2+4+8+⋯ ,1+2+4+8+\cdots,

and still diverges.

These are not contradictory statements.

They answer different questions.

The ordinary series asks:

do the partial sums settle towards a finite limit?

They do not.

Analytic continuation asks:

what larger analytic function is determined by the function that this series represents where the series does converge?

That function has value −1-1 at z=2z=2.

(The equation 1+z+z2+⋯=1/(1−z)1+z+z^2+\cdots=1/(1-z) is an equality of ordinary convergent sums only when ∣z∣<1|z|<1. Outside that circle, 1/(1−z)1/(1-z) is the continuation of the function, not the ordinary sum of the divergent geometric series.)

This distinction becomes extremely important when we meet the Riemann zeta function.

The famous minus one twelfth

Perhaps the most notorious example is

1+2+3+4+5+⋯ .1+2+3+4+5+\cdots.

You may have seen the claim

1+2+3+4+⋯=−112.1+2+3+4+\cdots=-\frac{1}{12}.

Written without qualification, that statement is deeply misleading.

In the ordinary sense of adding the terms,

1+2+3+4+⋯1+2+3+4+\cdots

does not equal a negative fraction.

Its partial sums are

1,3,6,10,15,…1,\qquad3,\qquad6,\qquad10,\qquad15,\qquad\ldots

and grow without bound.

The connection with −1/12-1/12 comes from the Riemann zeta function.

For

Re⁡(s)>1,\operatorname{Re}(s)>1,

zeta is represented by

ζ(s)=∑n=1∞1ns.\zeta(s)=\sum_{n=1}^{\infty}\frac{1}{n^s}.

If we formally substitute s=−1s=-1, the terms become

1+2+3+4+⋯ .1+2+3+4+\cdots.

But s=−1s=-1 lies far outside the region where that defining series converges.

The substitution is therefore not legitimate as an ordinary infinite sum.

Analytic continuation, however, extends the zeta function to s=−1s=-1.

And the continued function satisfies

ζ(−1)=−112.\zeta(-1)=-\frac{1}{12}.

So the precise statement is

ζ(−1)=−112,\zeta(-1)=-\frac{1}{12},

not

1+2+3+4+⋯=−1121+2+3+4+\cdots=-\frac{1}{12}

in the ordinary sense of convergence.

(In some regularisation frameworks people use abbreviated language such as “the sum is −1/12-1/12”. The mathematically important point is that this is not ordinary convergence. For zeta, −1/12-1/12 is the value of the analytically continued function at s=−1s=-1.)

That is not a pedantic distinction.

It is the whole idea.

Analytic continuation preserves the function.

It does not alter the definition of convergence.

Why zeta needs analytic continuation

This brings us directly to the mathematics behind the Riemann Hypothesis.

The Riemann zeta function begins with

ζ(s)=∑n=1∞n−s.\zeta(s)=\sum_{n=1}^{\infty}n^{-s}.

That infinite series converges directly when

Re⁡(s)>1.\operatorname{Re}(s)>1.

(Re⁡(s)\operatorname{Re}(s) means the real part of the complex number ss. So Re⁡(s)>1\operatorname{Re}(s)>1 means every point lying to the right of the vertical line with real coordinate 1 in the complex plane.)

But the non-trivial zeros involved in the Riemann Hypothesis lie in the critical strip,

0<Re⁡(s)<1.0<\operatorname{Re}(s)<1.

So there is an immediate problem.

The region we most want to study lies outside the region where the original series definition directly converges.

Without analytic continuation, the familiar zeta series would not give us access to the critical strip at all.

Analytic continuation carries the function into that region.

In fact, the continued zeta function is meromorphic across the complex plane, with a single simple pole at

s=1.s=1.

[1]

(Meromorphic means analytic except at isolated poles. A pole is a controlled kind of singularity where the function becomes unbounded.)

That continuation is what makes it meaningful to ask where zeta vanishes inside the critical strip.

ASCII FIGURE // Why analytic continuation matters to zeta

A horizontal schematic of the complex s-plane showing the original zeta series region to the right of real part 1, the critical strip between real parts 0 and 1, and analytic continuation carrying the zeta function into that strip.

The defining zeta series lives directly to the right of Re(s)=1. Analytic continuation gives access to the critical strip where the non-trivial zeros live.

This is why the phrase appears so often across Riemann Console.

It is not peripheral terminology.

It is part of the doorway into the problem.

For the longer story of what happens once zeta enters the complex plane, see What is the Riemann zeta function? and What is the Riemann xi function?.

Gamma does the same sort of thing

The gamma function gives us another particularly clear example.

Euler's integral

Γ(s)=∫0∞ts−1e−t dt\Gamma(s)=\int_0^\infty t^{s-1}e^{-t}\,dt

converges directly when

Re⁡(s)>0.\operatorname{Re}(s)>0.

But Gamma obeys the recurrence

Γ(s+1)=sΓ(s).\Gamma(s+1)=s\Gamma(s).

Rearrange it:

Γ(s)=Γ(s+1)s.\Gamma(s)=\frac{\Gamma(s+1)}{s}.

That relationship lets us move leftwards into regions where the original integral no longer directly converges.

We can continue Gamma analytically across most of the complex plane.

But we do not get everywhere for free.

Gamma has simple poles at

0,−1,−2,−3,…0,-1,-2,-3,\ldots

and no zeros. [2]

So once again:

the original formula has a restricted domain;

the function continues farther;

and genuine singularities remain.

(The integral and the recurrence are not rival definitions of two different Gamma functions. Where both apply they agree, and analytic continuation ties them into one function.)

Analytic continuation is not a magic eraser.

It cannot simply wish singularities away.

When continuation hits a real obstruction

The geometric example gave us a singularity at z=1z=1.

Gamma has poles at the non-positive integers.

Zeta has a pole at s=1s=1.

These are reminders that analytic continuation has limits.

A function may encounter an isolated pole.

It may encounter a branch point.

Its continuation may depend on the path taken around a singularity.

In more complicated cases it may possess a natural boundary beyond which analytic continuation cannot proceed at all.

(A natural boundary is stronger than one isolated bad point. It is a boundary across which the function cannot be analytically continued at any point, however cleverly we try to re-express it.)

So the right picture is not:

“every formula can be extended everywhere.”

It is:

“an analytic function can sometimes be continued beyond the domain of the representation in which we first encountered it, and the analytic structure determines that continuation wherever it exists.”

That sentence is less magical.

It is also much more powerful.

One function, several formulas

There is another lesson hiding here.

We often learn functions as though each function comes with one defining formula.

But important functions commonly have many representations.

A function may appear as an infinite series, an integral, an infinite product, a differential equation, a recurrence relation, a contour integral, or another analytic expression.

Different representations may work best in different regions.

This is not a defect.

It is often how a mathematical object reveals its full structure.

The gamma function can be approached through an integral, recurrence, limits and products.

The zeta function can be approached through its Dirichlet series, Euler product, integral representations, functional equation and continued forms.

The Riemann xi function then packages zeta with other analytic factors so that its central symmetry becomes especially clean.

A formula is therefore often less like an identity card and more like a viewpoint.

The function is the object that remains consistent as those viewpoints overlap.

A nineteenth-century change in what a function could be

Analytic continuation belongs to the great nineteenth-century development of complex analysis.

Augustin-Louis Cauchy, Bernhard Riemann and Karl Weierstrass approached analytic functions from somewhat different directions.

Cauchy's theory emphasised complex differentiation, integration and the powerful consequences of contour integrals.

Riemann developed a deeply geometric viewpoint.

Weierstrass systematically developed the power-series viewpoint, in which an analytic function can be understood through overlapping local series.

These approaches ultimately describe the same remarkable class of functions. [3]

The differences were not merely stylistic.

They changed what mathematicians understood a function to be.

Earlier mathematics often treated formulas and functions as nearly inseparable. During the nineteenth century the broader modern idea of a function as a correspondence became increasingly explicit.

Against that background, analytic functions stood out as a particularly rigid and structured species.

They were not arbitrary correspondences.

Their local behaviour constrained their global behaviour.

Analytic continuation became one of the clearest expressions of that rigidity.

(There was no single moment when one mathematician “invented analytic continuation” in its complete modern form. The modern theory grew through nineteenth-century complex analysis, especially the traditions associated with Cauchy, Riemann and Weierstrass.)

Weierstrass's stepping stones

The overlapping-disc picture we used earlier is particularly close to the Weierstrass viewpoint.

Begin with one convergent power series.

That is a local analytic element.

Choose a point inside its disc of convergence and expand the same function around the new centre.

If the new disc reaches farther, we have extended our knowledge of the function.

Repeat.

A chain of overlapping analytic elements can carry the function through a much larger domain. [3]

The crucial point is that each new step must agree with the preceding one on their overlap.

We do not get to redesign the function at each stage.

The local pieces belong to one analytically connected whole.

This is a beautiful reversal of everyday intuition.

Normally, knowing something locally tells us very little about what happens far away.

For analytic functions, local information can have global consequences.

What Riemann surfaces add

There is, however, a complication.

Sometimes travelling around the complex plane and returning to what appears to be the same point can lead us to a different branch of a function.

The complex logarithm is a classic example.

So is the square root.

This sounds as though uniqueness has failed.

It has not.

Along each local continuation, the analytic step remains rigid. The complication is global: the ordinary complex plane may not be a large enough domain on which to regard the whole continued object as single-valued.

Riemann's geometric viewpoint led to the use of surfaces on which different branches are kept on different sheets joined together in a controlled way.

A function that looked multi-valued on the ordinary complex plane can then become single-valued on a Riemann surface.

(Branching is not an arbitrary choice of values. It records the fact that continuation along different paths can reach different analytic branches when the domain has the right kind of hole or branch point. A Riemann surface keeps track of those branches geometrically.)

This is one of the places where analytic continuation stops being merely a technique and begins changing our idea of the space on which a function lives.

How do we know a continuation is really the same function?

Suppose we have one analytic expression f(z)f(z) on a region D1D_1, and another F(z)F(z) on a larger or neighbouring region D2D_2.

To claim that FF continues ff, the two must agree on a connected overlap where both are analytic.

If

f(z)=F(z)f(z)=F(z)

through an open part of

D1∩D2,D_1\cap D_2,

then the identity theorem tells us that the agreement is not accidental.

The two expressions represent the same analytic function throughout the connected overlap.

This is the mathematical lock that holds continuation together.

A proof of an analytic continuation therefore does not amount to saying:

“this new formula looks plausible outside the old domain.”

One establishes the new expression as analytic where claimed and proves that it agrees with the old function in a region where both are valid.

After that, uniqueness does the rest.

That is why continuation can legitimately carry a function into regions where its first formula has ceased to make sense.

The connection has been proved in the overlap.

Can two continuations disagree?

Locally, compatible analytic continuation is unique.

Globally, however, topology can matter.

If a domain contains holes or branch points, following different paths may lead to different branches.

This is not a contradiction with the identity theorem.

Along each sufficiently small overlapping step, continuation remains rigid.

The difference appears because the paths themselves can wind differently around singularities.

This phenomenon is called monodromy.

(If every closed path can be continuously shrunk to a point without crossing a singularity, the domain is called simply connected. In such settings, path-dependent branching is much easier to rule out.)

We do not need the full theory here.

The important point is simply that two ideas must be kept separate:

analytic continuation is locally unique;

the global analytic object may nevertheless require several sheets or branches.

This is precisely why complex analysis gave rise to Riemann surfaces.

Analytic continuation is not regularisation

Because divergent series attract dramatic examples, several related ideas are often mixed together.

Analytic continuation is one of them.

Regularisation is another.

Summability methods are another.

They are not interchangeable.

Analytic continuation begins with an analytic function already defined somewhere and extends that function while preserving analytic compatibility.

A regularisation procedure may assign a finite quantity to an expression that diverges in its ordinary interpretation.

Sometimes analytic continuation supplies that regularised quantity.

Sometimes a different method does.

The statement

ζ(−1)=−112\zeta(-1)=-\frac{1}{12}

is an analytic-continuation statement about zeta.

Turning that into the unqualified claim

1+2+3+4+⋯=−1121+2+3+4+\cdots=-\frac{1}{12}

suppresses the distinction between the continued function and the divergent series.

The shorter sentence may be memorable.

The longer explanation is the mathematics.

What analytic continuation does not do

It does not force a divergent series to converge.

It does not guarantee that every function can be extended.

It does not allow arbitrary values to be assigned beyond the original domain.

It does not remove genuine singularities merely because they are inconvenient.

It does not imply that one formula must work everywhere.

And it does not mean that a continuation suggested numerically is automatically proved.

What it does provide is a rigorous way to recognise when an analytic object extends beyond the range of the representation through which we first met it.

That is already extraordinary enough.

Why this matters beyond zeta

Analytic continuation appears throughout modern mathematics and mathematical physics.

Special functions are routinely first defined in one region and then extended.

Gamma is analytically continued beyond the half-plane where Euler's integral directly converges.

Zeta is continued beyond the half-plane where its defining Dirichlet series converges.

Hypergeometric functions, Bessel-related functions, exponential integrals and many other standard objects have continuation formulas controlling their behaviour across different parts of the complex plane.

The technique is also deeply connected with differential equations, spectral theory, number theory and the study of singularities.

Whenever an analytic object is first presented by a formula whose convergence is local, the question naturally arises:

is the formula's boundary the object's boundary?

Often, it is not.

A small glossary

Analytic function

A complex function with the strong regularity associated with complex differentiability. Locally, it can be represented by a convergent power series.

Analytic continuation

The extension of an analytic function beyond its original domain in a way that agrees analytically with the original function wherever the two descriptions overlap.

Domain

The set of inputs on which a function is being considered or defined.

Power series

An expression of the form

a0+a1(z−z0)+a2(z−z0)2+⋯a_0+a_1(z-z_0)+a_2(z-z_0)^2+\cdots

that represents an analytic function within its radius of convergence.

Identity theorem

A theorem expressing the rigidity of analytic functions: under the appropriate conditions, sufficiently rich agreement inside a connected domain forces two analytic functions to be identical there.

Singularity

A point at which an analytic function fails to behave regularly. Poles and branch points are examples.

Pole

An isolated singularity at which a function becomes unbounded in a controlled way.

Branch point

A point around which analytic continuation can move between different branches of a multi-valued analytic object.

Riemann surface

A surface constructed so that a function with several branches in the complex plane can be treated as a single-valued analytic function on the enlarged surface.

The journey to the Riemann Hypothesis

We can now see why analytic continuation belongs so naturally in the Riemann story.

Euler studies infinite sums such as

∑n=1∞1ns.\sum_{n=1}^{\infty}\frac{1}{n^s}.

For suitable values of ss, the series converges.

Euler also discovers that the same function can be expressed through a product over primes.

The primes are now inside the analytic object.

Riemann lets the variable ss move through the complex plane and uses analytic continuation to treat zeta beyond the half-plane where the original series converges. [4]

The critical strip becomes accessible.

The non-trivial zeros become objects that can be studied.

The functional equation reveals a reflection symmetry.

Gamma enters the completion.

The Riemann xi function packages that analytic structure into a cleaner entire function.

And the Riemann Hypothesis asks where its relevant zeros lie.

ASCII FIGURE // From a convergent series to the Riemann Hypothesis

A conceptual chain showing how Euler's convergent zeta series leads through analytic continuation to complex zeta, its non-trivial zeros, the completed xi function and finally the Riemann Hypothesis.

Analytic continuation is the bridge that lets the zeta function leave the region of its defining series and enter the part of the complex plane where the Riemann Hypothesis lives.

Without analytic continuation, that chain would break near the beginning.

The series would tell only part of the story.

The one sentence to remember

Analytic continuation is the rigorous process of extending an analytic function beyond the region where its original representation works, with the analytic information already known constraining the continuation uniquely wherever that continuation exists.

Or, more simply:

a formula can run out of road before the function does.

That is why

1+z+z2+⋯1+z+z^2+\cdots

can fail while

11−z\frac{1}{1-z}

continues.

It is why Euler's integral can stop directly converging while Gamma continues.

It is why the original zeta series can stop at Re⁡(s)=1\operatorname{Re}(s)=1 while the zeta function reaches into the critical strip.

And it is why one of the central ideas in the Riemann Hypothesis is not a new formula at all.

It is the recognition that the first formula we were given was never the whole function.

References

  1. NIST Digital Library of Mathematical Functions. “§25.2 Definition and Expansions — Riemann Zeta Function.” NIST Digital Library of Mathematical Functions, Release 1.2.8 (2026). Source
  2. NIST Digital Library of Mathematical Functions. “Chapter 5 — Gamma Function.” NIST Digital Library of Mathematical Functions, Release 1.2.8 (2026). Source
  3. Encyclopedia of Mathematics. “Analytic function.” Encyclopedia of Mathematics (2022). Source
  4. Bernhard Riemann. “Ueber die Anzahl der Primzahlen unter einer gegebenen Grösse (On the Number of Primes Less Than a Given Quantity).” Monatsberichte der Berliner Akademie / Clay Mathematics Institute manuscript collection (1859). Source

Glossary connections

  1. [..]Complex planeGLOSSARY · STANDARD MATHEMATICS
  2. [..]Critical stripGLOSSARY · RIEMANN HYPOTHESIS
  3. [..]Non-trivial zeroGLOSSARY · RIEMANN HYPOTHESIS